CN103294987A - Fingerprint matching method and fingerprint matching implementation mode - Google Patents
Fingerprint matching method and fingerprint matching implementation mode Download PDFInfo
- Publication number
- CN103294987A CN103294987A CN2012100535394A CN201210053539A CN103294987A CN 103294987 A CN103294987 A CN 103294987A CN 2012100535394 A CN2012100535394 A CN 2012100535394A CN 201210053539 A CN201210053539 A CN 201210053539A CN 103294987 A CN103294987 A CN 103294987A
- Authority
- CN
- China
- Prior art keywords
- image
- fingerprint
- coordinate system
- parameter
- information
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
- 238000000034 method Methods 0.000 title claims abstract description 162
- 230000008569 process Effects 0.000 claims description 39
- 238000006243 chemical reaction Methods 0.000 claims description 28
- 239000004744 fabric Substances 0.000 claims description 20
- 238000004458 analytical method Methods 0.000 claims description 14
- 238000012545 processing Methods 0.000 claims description 9
- 230000009467 reduction Effects 0.000 claims description 8
- 230000002708 enhancing effect Effects 0.000 claims description 5
- 230000011218 segmentation Effects 0.000 claims description 5
- 238000004088 simulation Methods 0.000 claims description 4
- 230000005428 wave function Effects 0.000 claims description 4
- 238000013213 extrapolation Methods 0.000 claims description 3
- 230000000737 periodic effect Effects 0.000 claims description 3
- 239000011148 porous material Substances 0.000 abstract description 62
- 230000005489 elastic deformation Effects 0.000 abstract description 12
- 238000001514 detection method Methods 0.000 abstract description 5
- 238000003672 processing method Methods 0.000 abstract description 3
- 238000012795 verification Methods 0.000 abstract description 2
- 210000004243 sweat Anatomy 0.000 abstract 1
- 230000008878 coupling Effects 0.000 description 37
- 238000010168 coupling process Methods 0.000 description 37
- 238000005859 coupling reaction Methods 0.000 description 37
- 238000005516 engineering process Methods 0.000 description 23
- 238000010586 diagram Methods 0.000 description 15
- 238000003860 storage Methods 0.000 description 15
- 230000009466 transformation Effects 0.000 description 15
- 238000000605 extraction Methods 0.000 description 12
- 230000008859 change Effects 0.000 description 11
- 239000000284 extract Substances 0.000 description 11
- 238000001914 filtration Methods 0.000 description 10
- 239000011159 matrix material Substances 0.000 description 10
- 238000009826 distribution Methods 0.000 description 9
- 238000005259 measurement Methods 0.000 description 9
- 230000006870 function Effects 0.000 description 8
- 230000008901 benefit Effects 0.000 description 6
- 238000012360 testing method Methods 0.000 description 5
- 230000008030 elimination Effects 0.000 description 4
- 238000003379 elimination reaction Methods 0.000 description 4
- 238000013178 mathematical model Methods 0.000 description 4
- 230000013011 mating Effects 0.000 description 4
- 210000003491 skin Anatomy 0.000 description 4
- 238000005728 strengthening Methods 0.000 description 4
- 238000013519 translation Methods 0.000 description 4
- 238000004422 calculation algorithm Methods 0.000 description 3
- 238000004590 computer program Methods 0.000 description 3
- 238000006073 displacement reaction Methods 0.000 description 3
- 230000000694 effects Effects 0.000 description 3
- 238000009432 framing Methods 0.000 description 3
- 238000010606 normalization Methods 0.000 description 3
- 238000010200 validation analysis Methods 0.000 description 3
- 238000012937 correction Methods 0.000 description 2
- 238000005520 cutting process Methods 0.000 description 2
- 230000007812 deficiency Effects 0.000 description 2
- 239000000428 dust Substances 0.000 description 2
- 238000009499 grossing Methods 0.000 description 2
- 230000001965 increasing effect Effects 0.000 description 2
- 230000004807 localization Effects 0.000 description 2
- 230000014759 maintenance of location Effects 0.000 description 2
- 230000000750 progressive effect Effects 0.000 description 2
- 230000009897 systematic effect Effects 0.000 description 2
- 230000007704 transition Effects 0.000 description 2
- PXFBZOLANLWPMH-UHFFFAOYSA-N 16-Epiaffinine Natural products C1C(C2=CC=CC=C2N2)=C2C(=O)CC2C(=CC)CN(C)C1C2CO PXFBZOLANLWPMH-UHFFFAOYSA-N 0.000 description 1
- 206010013786 Dry skin Diseases 0.000 description 1
- 238000007476 Maximum Likelihood Methods 0.000 description 1
- 235000012364 Peperomia pellucida Nutrition 0.000 description 1
- 240000007711 Peperomia pellucida Species 0.000 description 1
- 206010038743 Restlessness Diseases 0.000 description 1
- 230000003044 adaptive effect Effects 0.000 description 1
- 239000000654 additive Substances 0.000 description 1
- 230000000996 additive effect Effects 0.000 description 1
- 238000004364 calculation method Methods 0.000 description 1
- 230000019771 cognition Effects 0.000 description 1
- 239000000470 constituent Substances 0.000 description 1
- 238000013144 data compression Methods 0.000 description 1
- 238000000354 decomposition reaction Methods 0.000 description 1
- 230000007423 decrease Effects 0.000 description 1
- 238000013461 design Methods 0.000 description 1
- 230000004069 differentiation Effects 0.000 description 1
- 238000005315 distribution function Methods 0.000 description 1
- 230000037336 dry skin Effects 0.000 description 1
- 235000013399 edible fruits Nutrition 0.000 description 1
- 230000007613 environmental effect Effects 0.000 description 1
- 210000002615 epidermis Anatomy 0.000 description 1
- 238000012854 evaluation process Methods 0.000 description 1
- 238000009415 formwork Methods 0.000 description 1
- 239000011521 glass Substances 0.000 description 1
- 238000007417 hierarchical cluster analysis Methods 0.000 description 1
- 238000003709 image segmentation Methods 0.000 description 1
- 230000006872 improvement Effects 0.000 description 1
- 230000005764 inhibitory process Effects 0.000 description 1
- 230000001788 irregular Effects 0.000 description 1
- 238000003064 k means clustering Methods 0.000 description 1
- 238000004519 manufacturing process Methods 0.000 description 1
- 238000013507 mapping Methods 0.000 description 1
- 239000003550 marker Substances 0.000 description 1
- 239000000203 mixture Substances 0.000 description 1
- 238000003909 pattern recognition Methods 0.000 description 1
- 238000012805 post-processing Methods 0.000 description 1
- 238000011946 reduction process Methods 0.000 description 1
- 230000002787 reinforcement Effects 0.000 description 1
- 230000036555 skin type Effects 0.000 description 1
- 238000001228 spectrum Methods 0.000 description 1
- 230000001052 transient effect Effects 0.000 description 1
- 230000004304 visual acuity Effects 0.000 description 1
- XLYOFNOQVPJJNP-UHFFFAOYSA-N water Substances O XLYOFNOQVPJJNP-UHFFFAOYSA-N 0.000 description 1
Images
Landscapes
- Collating Specific Patterns (AREA)
Abstract
The invention relates to a fingerprint identification and verification method. Specifically, a fingerprint matching method and a fingerprint matching implementation mode comprise that fingerprints are turned into a standardized frame ( with unified regulations), wherein the frame is used for inputting, matching and storing a fingerprint template. The fingerprint matching method and the fingerprint matching implementation mode are an image processing method, can remove liner deformation and elastic deformation in an image, and turn a group of two-dimension digital images into a structural framework with the unified regulations. The invention further provides an image processing method for fingerprint image matching. According to the fingerprint matching method and the fingerprint matching implementation mode, the structure of the typical framework is involved, detection of sweat pore characteristics from the fingerprint image is involved, nonlinearity calibration of the elastic deformation is involved, an input program is involved, and a matching program is involved.
Description
Technical field
The invention relates to fingerprint recognition and verification method.Specifically, the present invention becomes fingerprint into standardized framework (a kind of framework with unified standard), and this framework is used for typing, coupling and storage fingerprint template.
Technical background
The feature of fingerprint is its level and smooth fingerprint ridge line and valley line, it be characterized by the position of these crestal line valley lines, at interval, shape and small details.Details is tail end and the bifurcated of crestal line.
Traditionally, fingerprint is adopted widely in the biometric field.Since 20th century in early days, people have understood the structure of fingerprint and difference (referring to Handbook of Fingerprint Recognition, D.Maltoni, et al, Springer 2003).The fingerprint science based on three basic criterions:
1. for different fingerprints, crestal line and the rill of individual epidermis take on a different character.
2. the structure type of fingerprint is different because of the difference of individuality, but they change in certain scope, therefore they can be carried out the classification of system.
3. the minor detail of the structure of fingerprint and single crestal line and rill changes never.
Consider above-mentioned criterion, many technology are invented, and these technology employing fingerprint information are carried out personal identification identification, and personal identification can be identified by the database that search includes fingerprint image (template).
Many automatic fingerprint matching algorithm application specific details information determine that whether two fingerprints are from same finger.Some technology are used other fingerprint ridge line feature (for example crestal line position, crestal line spacing and crestal line shape etc.).
In short, fingerprint authentication/identification comprises two stages: (1) typing; (2) coupling.
In the typing stage, computer programs process people's fingerprint image, and to the template of converting.These templates are carried out related with the metadata of personally identifiable information subsequently and are stored in the database.In the typing process, the fingerprint of acquisition is stored in the template database, only stores the fingerprint characteristic that those extract and show, can distinguish with certain form in this database.
Under the Validation Mode of matching stage (1:1 coupling), a people's fingerprint image will mate with the Template Information of a known identities, yet under recognition mode (1:N coupling), a people's fingerprint image will mate with all templates or one of them subclass of storing in the database.For the contrast of the template of each test fingerprint and storage, the mark of a coupling will calculate in system.
Therefore, emerging test fingerprint can contrast one group of template of storing and return the mark of coupling.Because test fingerprint need compare with the template of each storage, so it need be converted to the representation identical with template.Then, system will return a mark according to the similarity degree of (test) fingerprint that occurs and each template.If fractional value is enough high, be higher than user-defined threshold value, then system is claimed as coupling with it.
Analyze in different grades, the fingerprint models show goes out dissimilar features:
1. in first grade, the direction of fingerprint ridge line has constituted special model structure, and we can broadly be divided into these structures left-handed, dextrorotation, screw thread, arch and askew arch shape.(tented?arch)。These features are not enough to carry out accurate fingerprint matching, but they are still useful in carrying out fingerprint image classification and index.
2. second grade, the geometric position of each minor detail on fingerprint is extracted out, and the mutual relationship between second grade details can be given each fingerprint with individual character.
3. at Three Estate, the details in the fingerprint ridge line is detected, and these details are position, shape and branches of pore, and these information all are that differentiation is highly arranged.
The analysis of Three Estate often is applied in forensic science by the fingerprint expert artificially, and when the reliable finger print information that can only obtain part, and second layer data (small details) are not enough to make the finger print data that deterministic coupling just can be used this layer.Description of Related Art
Fingerprint recognition system can be divided into generally based on detailed information and based on correlativity.
Method based on minor detail at first will search out minutiae point, draws their relative positions on fingerprint then.When the consistance of confirming between details, include rotation, Pan and Zoom equally can be identified in interior global change.
Based on the method employing fingerprint crestal line of correlativity and the world model of rill, and calculate a mark based on the result of its correlativity.Transition matrix can be determined by calculating the peaked mode of related coefficient.
Fig. 2 is the process flow diagram of the general treatment step of previous legacy system.
In the first step, according to the difference of using, the marking or the live finger played up from ink obtain fingerprint image.In a single day system obtains fingerprint graph and it is stored in internal memory or the hard disk, often these images is strengthened and handles the quality that strengthens the fingerprint ridge line pattern.This generally includes contrast enhancing, noise elimination, filtering and smoothing processing.Some previous systems also can extract foreground information (as extract the crestal line figure from background) in this step.In the 3rd step, image correlation method or feature extraction are handled and will be employed.
Fig. 3 be " Adaptive flow orientation based feature extraction in fingerprint images ", Journal of Pattern recognition, Vol.28, no 11, pp1657-1672Nov.1995and in US Patent 6, the process flow diagram of a kind of Feature Extraction Technology that is widely adopted that proposes in 049,621 one literary composition.
At first, image is carried out piecemeal and assess out the position of every topmost fingerprint ridge line.Application prospect/background segment technology is separated the fingerprint part of image from background then.Next, using the bianry image technology extracts crestal line feature (being labeled as 1) from non-crestal line feature (being labeled as 0).The width of these crestal line features is usually more than a pixel and may comprise noise (noisy artefacts).In level and smooth step, these noises will be removed, and long structure is carried out smoothing processing.In next step, it is wide that the fingerprint ridge line structure after level and smooth will be reduced to a pixel.From the fingerprint ridge line structure after the reduction, extract position and the direction of minor detail feature then.Some system applies are removed aftertreatment (cleanup post-processing) step and are removed false minutia.
The 4th of coupling flow process goes on foot normally calibration steps.System applies details position or crossing dependency information before most of are determined global affine transformation, and eliminate comprising translation and being rotated in the conversion of interior geometric position between query fingerprints and template fingerprint.
Systems before some, No. 6049621 United States Patent (USP)s have for example proposed a kind ofly can set up conforming method between the meeting point of two images.This method at first need find the point of a pair of correspondence at least between query image and template image, and puts conduct with reference to point with this.Curve (fingerprint ridge line after the reduction) according to reference point locations information and reference point place calculates conversion and rotation parameter between corresponding crestal line.Next, the right index of all possible details between application reference dot generation query fingerprints and template fingerprint.Then using iterative handle to determine each details to transformation relation (translation and rotation).System will for every pair of possible details to calculating a calibration mark, and the details that mark is higher than predetermined value is to being claimed as corresponding details pairing.The advantage of said method is that it can only not calculate global change's (translation, rotation and scale transformation), and can handle local elastic deformation by calculating local transformation parameter.Yet the quality that the success ratio of this method relies on feature extraction and crestal line to survey.
WO2005/022446 proposes pore and gets the uniqueness that position, shape and distribution have height, and these features can separately or be used for carrying out identification or the checking of fingerprint as the side information of detailed information.
The shortcoming of system before
The coupling of password or identification code and another password or identification code relates to the contrast to two completely specified parameters, and this can make accurate coupling become easy.On the other hand, finger print matching system or other biological measuring system comprise the contrast to the high complexity function, and this can make this system more complicated.
The mistake of two types can appear in the measurement in the biometric system and coupling: the coupling of mistake and wrong not coupling.This not exclusively accurate coupling has very big variation mainly due to same finger at different images, and these change normally because displacement, rotation, part are overlapping, influence, noise and the feature extraction mistake of elastic deformation, different pressure, different skin conditions, light cause.When the fingerprint of same finger looks that fingerprints different or different fingers look very similar, reliably fingerprint image is mated the problem that has become a difficulty.
The inconsistent of the minutiae point that extracts between query fingerprints and template fingerprint and crestal line structure caused by multiple phenomenon, comprising:
1. the finger scan of query fingerprints and template fingerprint is regional different.
2. because the contacting of the nonrepeatability of manual operations or the unexpected finger that causes and scanner, the structure that these contacts can temporary transient change fingerprint ridge line.
3. different skin and environmental factor low contrast and/or the picture noise that can cause image, these factors comprise: dry skin, perspiration, dust, humidity, at the glass face of scanning device residue etc. is arranged.
4. in feature extraction, crestal line detection and reduction process, can there be mistake and culture noise.
The inconsistent meeting of these mistakes and feature extraction and crestal line detection then is transmitted to calibration process, and this will finally influence correct conforming foundation.
For example, in the method based on details, lower or data are relative with overlapping region between template hour when the quality of fingerprint image, and it is difficult extracting consistent minutiae point.The extraction of inconsistent minutiae point will influence determining of conforming foundation and transformation matrix.
Technology based on correlativity is easy to generate noise, cut, pollution, and the height that assesses the cost.
Therefore, the result that handles of fingerprint matching is usually because the elastic deformation of variation, fingerprint positions and the direction of the noise during Image Acquisition, contrast different and the image that causes owing to main body application of force angle difference makes matching result imperfect usually.
Therefore the finger print matching system before has comprised many steps usually and finished coupling: (1) carries out pre-service to increase the contrast of crestal line structure to the image that obtains; (2) from the fingerprint ridge line that has strengthened, extract the feature of discrimination; (3) after calibration process, the details of image and the details of template are contrasted one by one.In fact, in the system formerly, template may only comprise the information of position, type (bifurcated or end) and details direction.Before the coupling mark calculates, to mating all these three execution that step all can repeat each time.
As mentioned above, system of fingerprints comprises calibration process usually, and this was handled before coupling and carries out, and guaranteed that image is calibrated or adjusted, and two width of cloth images is accurately compared being easy to.Calibration process combines conversion and rotation usually, and the two has constituted the conversion parameter of definition overall situation calibration together.Calibration steps before depends on minor detail information, and this will inevitably cause by the comprehensive problem that produces of two X factors (transition matrix between the consistance between minutiae point and the minutiae point set).The calculating of conversion parameter depends on the consistance of minutiae point, and conforming foundation depends on the accurate assessment to conversion parameter again.Any mistake that occurs in any appraisal procedure all will be passed to the accuracy that follow-up coupling step was handled and will be reduced to the back level.
And because the calculating of the conversion parameter of previous system depends on the contrast to query fingerprints and template fingerprint, processing procedure need be carried out mode of comparing to pursue.For example, the carrying out that calibration process need repetition when at every turn query fingerprints and template fingerprint being compared.In fingerprint recognition was used, when system attempted to find out the ownership of query fingerprints from big database, calibration operation need be repeated repeatedly up to the identity of determining query fingerprints.
Another deficiency of previous system is their only partial informations in the employing fingerprint image, mainly is crestal line pattern and minor detail information.Because consistent detection and set up the difficulty of pore correlativity, Automated Fingerprint Identification System before can be ignored these permanent, constant specific characteristics.The feature of Three Estate (pore) generally be in forensic science by the fingerprint expert artificial cognition because in this case some the time only might obtain the part fingerprint image and secondary characteristics be not enough to identify.For some specific skin type or mechanic, the negligible amounts of the minutiae point that can be arrived by reliable detection, this can lead to the failure in the typing stage usually.
The adding of pore feature is the feature-rich space greatly, when itself and detailed information are replenished mutually, can reduce wrong typing rate and increase matched accuracy.Although yet the 3rd layer analysis has increased accuracy rate to a great extent, this analysis also needs high-resolution scanning.In view of the density height of pore, consistent feature extraction and the correlativity between fingerprint are difficult to obtain.The needs of high resolution scanning (more than the 1000dpi) have hindered being extensive use of of third level analysis.Yet, because the lifting of the high resolving power fingerprint image equipment and technology among the decline of device fabrication cost and the picture WO2005/022446, people have been again to having produced interest with pore information as the side information of minutia, can improve the accuracy of system like this and the availability of finger print information in non-judicial the application.
Therefore, because it is mentioned above, one can overcome before the method for deficiency of system be necessary, similar variation can be eliminated or reduce to this method in a normalized structural framing, and pore information can be combined to strengthen fiduciary level and the accuracy of fingerprint recognition with detailed information.
Summary of the invention
The present invention is a kind of image process method, and it can be removed linearity in the image and elastic deformation and one group of two-dimensional digital image is changed into a structural framing with unified standard.The present invention has proposed the image processing method of object fingerprint images match equally.
One aspect of the present invention relates to the structure of this typical framework.Another aspect relates to and survey the pore feature from fingerprint image.Also have an aspect to relate to elastically-deformable gamma correction.Also have an aspect to relate to recording program.Last aspect is about matcher.
In a main implementation of the present invention, a method of handling fingerprint image has been proposed.This method has comprised following steps: for each calibration of wanting processed image, this calibration comprises conversion and/or rotation; Image is divided into some zones, and each image-region is measured in the following parameter at least one: backbone line position, average ridge distance between centers of tracks and phase place, and above-mentioned measured value stored; For each pending image, above-mentioned each regional measured value is projected in first coordinate system of a multidimensional and represent this image with the first above-mentioned coordinate system, in this coordinate system, the distance of the relevant parameter of two width of cloth images has indicated the dissimilarity of two width of cloth images.
Calibration process comprises following step: the biological center and biological axle of determining fingerprint in the image; A common reference point and axis of reference are set; Mobile image overlaps biological center with reference point, image rotating is consistent with the axis of reference direction with the biology axle.Do not appear in the image if obtain the biological center of fingerprint, should be in conjunction with extrapolation and the known fingerprint image of existing fingerprint part crestal line are estimated the position of biological center outside image.
Can come parts of images in the simulated domain as model with periodic function (in general being the sin curve), in the image, above-mentioned parameter can be measured in above-mentioned model image and/or above-mentioned true picture in this section.Estimation error can be calculated, and estimation error is in the parameter that model is measured and the difference of measuring parameter in non-modeled image.If the measuring error in a certain zone has surpassed preset threshold, just need to use further image segmentation step that Region Segmentation is become subregion, then subregion is carried out parameter measurement.
Image can be in first coordinate system represents that with the vectorial V corresponding with the parameter value of measuring in each zone of each image this coordinate system has constituted a vector space.The difference of above-mentioned representative distance or visibility can strengthen with some technology.In an implementation of the present invention, these measurement data are mapped in second coordinate system representative of the expression information of two width of cloth images in this coordinate system distance big than in first coordinate system.In another implementation of invention, the method for application examples such as pivot analysis dimensionality reductions such as (PCA) strengthens otherness or observability, can eliminate a dimension so at least from first coordinate system.The application characteristic value/proper vector of defencing jointly in the poor Matrix C in master pattern set decomposes to obtain eigenvalue V ', here pattern mean vector M=E[V], C=E[(V-M) { (V-M) } ^T].
System can come for strengthening discrepancy score of system assignment by the distance of the representative between expression information, and this discrepancy score is signifying to represent the dissimilarity of information.
In the implementation of invention, can the part dimension of coordinate system not handled, it is processed to have only the dimension K of the non-eliminating of sub-fraction to need in all dimensions, and the dimension of eliminating is that those mark variations that cause after with it elimination are less than the dimension of a predetermined second threshold value.
Image can be classified according to the position of corresponding expression information in coordinate system.All images in a certain specific image classification needs to determine a class template image that this image is the average pattern M_c of such c.Every width of cloth image in the class all can be divided into some zones, and the size in zone is based on its distance to core.
If the size of subregion is less than predetermined second threshold value, then should the zone or subregion no longer divided, directly use transformation parameter and not to further cutting apart.
Transformation parameter for each area applications in the piece image can be determined, the parameter of one width of cloth candidate image represents and will represent to compare with the corresponding parameters of the same area of class template image M _ c, and candidate image is represented and the distance of template image between representing is defined as regional conversion parameter.
Represent the similarity degree of information and the corresponding region representation information of template image according to candidate image area, system can be that each contrast distributes a mark.If this mark equals or exceeds the 3rd a predetermined threshold value, then this conversion parameter is composed to the candidate image respective regions, use this conversion parameter expression information is changed.If mark is lower than the 3rd a predetermined threshold value, then should further be divided into subregion in the zone, and the parameter of subregion will be measured, again comparison.
System may obtain more images, and correspondingly, the data of storage and coordinate system also can upgrade.System identifies the position of details and direction in can the image after the calibration of every width of cloth and with it storage, these data will comprise position, shape and the size of pore at least.Pore information and/or detail data can be projected onto in the coordinate system, and coordinate system can correspondingly comprise these data to come in to upgrade.
System will be divided into cluster with expression by using clustering technique, and this sorting technique may be the technology of k-means cluster or an iteration cluster, represent that here the classification of information will be according to the representative distance between expression information and existing group relative expansion.
In an implementation of invention, by mating recently determining candidate image and memory image.This has further comprised following steps: obtain candidate image; Measure and store the parameter of candidate image; Measured value projected in first coordinate system and with coordinate system upgrade; First width of cloth image applications classified, cuts apart, changes determine, steps such as scoring and conversion; Determine the position of details in the candidate image and direction and it is projected first coordinate system; Give a possibility mark to candidate image, this possibility mark is the possibility that candidate image is included into the predetermined image class; According to the possibility mark of above-mentioned candidate image to each image class it is included in one or more image class; The expression information of the detail data of template image in contrast candidate image and the same item; Identification pore data are also with it storage in candidate image, and these data comprise position, the shape and size of pore at least; The expression of the pore data of template image in contrast candidate image and the same item; Distribute the coupling mark, mate mark here and distribute according to the similarity degree of the respective regions of template image in candidate image area and the same item; If the coupling mark is higher than predetermined threshold value, the statement coupling.The possibility allocation step can be further divided into to be represented candidate image the contrast of information and predetermined class average and launches to determine that according to the expression information in average and the class image is included into these two steps of possibility of certain class.The result shows that candidate image does not meet any predetermined class when estimation, and system will state non-matching.
In another main implementation of invention, a kind of implementation of handling fingerprint image has been proposed, this implementation comprises: by the method for changing and rotation is calibrated and image is divided into the zone image, go out the method for at least one following parameter with each area measure to every width of cloth image: the direction of backbone line, the average ridge distance between centers of tracks, phase place, store the method for above-mentioned measured value, above-mentioned each regional measured value is projected the method for first coordinate system of a multidimensional, the method of presentation video in above-mentioned first coordinate system, the representative distance of the relevant parameter by two images indicates the dissimilar degree methods of respective image.This implementation comprises equally: determine biological center and biological axis in the fingerprint image; Set general reference point and general axis of reference; Converted image places general reference point with the biological central point of fingerprint; Rotating image is with method such as the biological axis direction of fingerprint is consistent with the axis of reference direction.
When the biological center of fingerprint did not have in image, this implementation can be estimated the image position at the biological center of fingerprint outward in conjunction with extrapolation and the known fingerprint pattern of conventional images fingerprint pattern crestal line.
In an implementation, this implementation includes modeled method, this method is used as the parts of images in the sin period of a function wave function modeling zone, and here, above-mentioned parameter is measured in above-mentioned model image and/or non-modeled image.
This implementation also can be determined an evaluated error, and this evaluated error is the difference of the parameter that obtained by model measurement and the parameter that obtained by non-modelling image measurement.
This implementation also comprises second method cut apart of step, here, if the evaluated error of specific region is surpassed predetermined first threshold, uses the method that second step cut apart so Region Segmentation is become subregion, and subregion is carried out parameter measurement.
In an implementation of the present invention, progressive ground has comprised in above-mentioned first coordinate system with the method for vectorial V representative image in the implementation, and V is corresponding with the parameter value that each area measure of every width of cloth image goes out, and coordinate system constitutes vector space.
This implementation has also comprised the method for the observability that strengthens above-mentioned representative distance, and this method is that the parameter that will record projects in second coordinate system, and two width of cloth images in second coordinate system represent that the representative distance of information is bigger than first coordinate system.
The method of another enhancing is to reduce the dimension of first coordinate system, can use at least one dimension of subduing first coordinate system from the pivot analysis that proper vector/eigenwert decomposition obtains here.
This implementation also needs to comprise a discrepancy score distribution method, after each strengthens step, be discrepancy score of each enhancing system assignment according to the representative between expression information distance, this discrepancy score is used for indicating dissimilar degree between expression.
Advantageously, this implementation has comprised the part dimension elimination methods from handle with coordinate system, like this, only have the K of the non-removing property of sub-fraction can be processed in all dimensions, the dimension that is excluded be the dimension that the discrepancy score that causes after those are subdued can not be higher than the second predetermined threshold value.
Another implementation of the present invention has comprised the position difference according to corresponding expression information in the coordinate system, the method that image is classified, with comprise the method that the image of a certain kinds is classified, determine the class template image, this image is average image M _ c of class c.An implementation of the present invention has comprised each image in the class has been divided into some regional methods, and the size in zone is based on the distance of this zone to core.If the size of the subregion in a certain zone is lower than the second predetermined threshold value, this implementation will can not become this Region Segmentation subregion and directly use this regional conversion parameter.This implementation comprises the method for determining conversion parameter equally, this method can be used in the parameter of candidate image to be represented with class template image M _ c in the contrast of the expression of the relevant parameter of same area, and definite candidate image represents whether the representative distance between information and template image is regional conversion parameter.
The present invention is included as the method that contrast of each expression information distributes mark equally, and this mark is that the similarity degree according to the expression information of candidate image area and class template image respective regions distributes.
When mark was equal to or higher than the 3rd predetermined threshold value, this implementation can be composed conversion parameter this zone to candidate image, and used conversion parameter expression information is changed.When mark is lower than the 3rd predetermined threshold value, the parameter of subregion will further be cut apart and measure to implementation to this zone.
In another implementation of the present invention, have method that image obtains and obtain more images and correspondingly data and the coordinate system of updated stored.This can comprise a details localization method and come the position of definite details in the fingerprint image that each calibrate and direction and these information are stored.This also can comprise, and a pore localization method determines the related data of pore in the fingerprint image that every width of cloth was calibrated and with it storage, these data will comprise position, the shape and size of pore at least.This implementation also can comprise and pore and/or detail data projected coordinate system and upgrade the method that coordinate system makes it to comprise these expression data.
Another aspect of the present invention is to comprise a clustering method by using clustering techniques such as k-means expression information to be classified.Another kind of clustering technique is the iteration clustering technique, by expression information itself and existing faciation the representative distance of launching is classified to expression information here.
An implementation of the present invention has also comprised matching process, and this method is used for the image of candidate's image and storage is compared to determine whether image mates.This method can be applied to: obtain candidate image, measure the parameter of candidate image and with it storage, project the value that measures in first coordinate system and upgrade first coordinate system.Matching process can comprise following method: classify, cut apart, conversion is determined, scoring, conversion candidate image; Determine position and the directional information of details in the candidate image, above-mentioned information is projected first coordinate system; Distribute the possibility scoring to candidate image, this mark is the possibility that candidate image is included into certain predetermined class; By the possibility scoring one or more image classes are included in above-mentioned candidate image; The expression information of candidate image in similar and template image detail data is compared; Determine the pore data of candidate image also with it storage, these data comprise position, the shape and size of pore at least; The expression information of the pore data of candidate image in similar and template image is compared; Coupling allocation scores step, represent that with similar middle template image respective regions the similarity degree of information is mark of a coupling of contrast distribution of each expression information according to the expression information of candidate image area here, if the coupling mark is higher than a predetermined threshold value, then be claimed as coupling.In this implementation, the possibility allocation scores may also can comprise following method: the expression information of candidate image and the equal value information of the class of being scheduled to are compared; And according to the class average and the expression information the expansion evaluate image whether belong to such.Appraisal procedure can comprise following method: according to the possibility assessment, when candidate image does not belong to predetermined class, state non-matching.
In another main implementation, computer program contains one and is included in this instruction using method that is described at interior readable media.
More concrete going up said, constitutes a typical architecture and comprises following steps: input picture is divided into a chunk (zone); The application parameter modeling technique carries out modeling to interested feature in the piece, specific to fingerprint image, direction and the spacing of fingerprint ridge line in the piece is carried out modeling; Determine the center of image inside and the direction of image; According to center and the direction of image, calibrating direction and spacing model; By with in direction and spacing model transferring to the new coordinate system to reduce its dimension; Vector behind the dimensionality reduction is projected in this coordinate system.
Another aspect of the present invention is that it provides a method of extracting pore information.This method has comprised following steps: determine possible pore position; Local density's information around candidate's pore is carried out modeling; According to local density's information, they remove false pore information to the relative distance of contiguous crestal line and the direction of contiguous crestal line.
Another aspect of the present invention has provided elastically-deformable method between a kind of removal two width of cloth images.This method has comprised step: image is divided into one group of regional area; Conversion parameter between estimation data (query image) and target (template image); Conversion parameter is applied to each zone of query image, and obtains calibration error; If error amount is sufficiently big, this just means and still has elastic deformation in the zone, this zone is divided into one group of littler zone again, and each littler zone is assessed again; It is sufficiently low that this evaluation process is performed until each regional error amount; The transformation parameter final to the corresponding area applications of query image changes into formwork structure with image.
Another aspect of the present invention is to have proposed a fingerprint image input system, and this system comprises following steps: obtain fingerprint image and the crestal line structure is carried out modeling; Model parameter is projected in the typical structural framing; From image, extract the minor detail set; Extract pore information; Detailed information is projected in the typical framed structure; The structure template is also used it storage for following matching process; Based on their distance in the typical feature space template is classified.
Another aspect of the present invention has provided a kind of method of determining the fingerprint of inquiry from the template of one or more storages.This method has comprised following step: the fingerprint image that obtains inquiry; The crestal line structure is carried out modeling; Determine inner center and direction; Model parameter is projected in the typical feature space; The fingerprint image that calculates inquiry belongs to the possibility of certain class template; Elastic deformation between the average of the template under estimation query fingerprints and each its possibility; The query fingerprints image applications overall situation or local deformation are changed into the average of corresponding template with it; Extract the detailed information of normalization; Compare by the detailed information to each template of storing in the detailed information of normalization and the class and to determine details coupling mark; Generate an overall mark by possibility mark and details coupling mark; By being compared, overall mark and predetermined value make the decision of whether mating.
Description of drawings
Some other aspect, feature and advantage above-mentioned and invention can be understood by following specifying more comprehensively of first-selected implementation to invention, yet this is not limitation of the present invention, and it just is used for making an explanation.
Fig. 1 is the sketch of the digitizing example of inside center (core), direction, crestal line, minor detail and a pore of describing fingerprint image.
Fig. 2 is the process flow diagram of previous conventional fingerprint matching system step.
Fig. 3 is the process flow diagram of previous traditional characteristic extracting method step.
Fig. 4 describes to use multiresolution method with the sketch of the implementation of fingerprint image lattice.
Fig. 5 is the process flow diagram of using the implementation step of a typical structure framework of method construct of the present invention.
Fig. 6 uses method of the present invention to the process flow diagram of the step of the implementation of pore information extraction.
Fig. 7 uses the process flow diagram that method of the present invention is carried out the step of the implementation that gamma correction handles.
Fig. 8 uses method of the present invention is carried out the step of typing to finger print information process flow diagram.
Fig. 9 uses method of the present invention to the process flow diagram of the step of the implementation of the template identification query image of one or more storages.
The process flow diagram of Figure 10 finger print matching system step of the present invention.
Embodiment
The specific descriptions of the implementation of first-selection given below and diagram will make the present invention more clearly be understood.
In the following description, function and the operational character that it is well known that are not done too much introduction, to avoid causing reader's bluring inventive concept because of unnecessary details.
Following part, we will introduce the method for processing fingerprint image and a kind of implementation wherein, this mode changes into point on the feature space with template, these put relative fingerprint outward appearance is constant: this space refers to a kind of typical expression, means that fingerprint image is arranged in standardized space.And feature space represents to improve data compression rate and to the strong robustness of the noise (for example cut) of the image that obtains.We can introduce one equally to the non-linear extension in this space, and this makes us to handle the elastic deformation of the fingerprint that occurs.What ensuing article was described is to use this characteristic feature space to finish the image processing.
A very important advantage of the present invention is that it is that specialized application is on crestal line structural model (local direction of crestal line and spacing).When converting fingerprint to typical form, second level crestal line feature (minor detail) and third level fingerprint characteristic (pore) are dispensable.This typical expression information is independent of the second layer and the 3rd layer of feature.This also needs the problem that this is faced a difficult choice is estimated in calibration when needs are set up the correlativity of above-mentioned feature of position and test pattern and template with regard to having been avoided.
More progressive is, in special implementation of the present invention, feature space is broken down into ground floor finger print information (as arch and screw thread) easily, and this has just reduced the complexity of the system-computed of 1:N type search.
The structure in characteristic feature space
An implementation of the present invention is process flow diagram as shown in Figure 5, the figure illustrates the step of the typical framework of structure.Discovering before, crestal line pattern have only limited kind of topological structure, for example left-handed, dextrorotation, screw thread, arch and account bow.Feature space can constitute under a kind of pattern of off line, for example handles one group of fingerprint image that pre-stores and constructs initial feature space; Also can under online pattern, constitute gradually structural attitude space in the fingerprint typing and when handling.In both cases, processing begins in the mode that the application parameter model represents fingerprint pattern.
According to an implementation of the present invention, after from the template of scanning device or storage, obtaining fingerprint image, a graduation lattice technology is used to image is divided into one group of region unit (step 502), and then the crestal line that is used to every of parameter model partly carries out modeling (step 503).Though can select any method easily that image is cut apart, in general region unit is foursquare.
For each piece of image, the estimation of the characteristic parameter of piece intercycle fingerprint pattern is obtained by pattern being modeled to the proper model function.In the implementation of a first-selection of the present invention, adopt the sin model to represent regional fingerprint characteristic, although other periodic wave function also can be used as mathematical model.By mathematical model is applied on the image, just can be to obtaining the estimation to pattern parameter the corresponding measurement of model.The direction of mathematical model and frequency can be used many other technologies and estimate (step 504).Therefore, in the implementation of a first-selection, each region unit is transformed into frequency field, so just can comes direction and the frequency of estimating signal by the peak value of locating main wave spectrum.The sin model that then application is estimated frequency and direction parameter carries out comprehensively the fingerprint ridge line model in the region unit.In the implementation of a first-selection, by the inner product of computational data and unified model, the phase place of crestal line part also can be estimated goes out.
In an implementation of the present invention, can calculate estimation error by contrast integrated data and True Data.This error can change with regional change, and this is the result of inconsistent influence, for example displacement, rotate, overlap, elastic deformation, pressure difference, skin condition difference, light influence, noise and feature extraction mistake, these were all being mentioned before.System can carry out the estimation of estimation error to each zone: the error amount in the zone is higher than predetermine level, and this just means that the data too complex in the zone can't estimate with current model.When this happens, this zone can further be divided into one group of littler zone.The execution that step 504 can be repeated is lower than predetermined grade up to the simulation error that All Ranges all simulated and each is regional of image.
Step 505 and 506 is centralized positioning and calibration steps: they are used for reorientating the biological center of fingerprint and recalibrating the direction of fingerprint with reference axis with the geometric center of image.Describing these among Fig. 5 can occur in after the step 504, but these steps also may be carried out before step 504.The core of fingerprint (inside center) is to be positioned at the innermost zone of crestal line.In general, it is positioned at the centre of fingerprint, but because the zone of scanning, he may not be positioned at the center of image, even does not appear on the image.When core appears on the image, can use a lot of other technologies it is detected.In the implementation of a first-selection of the present invention, determine position and the direction (step 505) of core with one group of circular symmetric function.Implementation is with by two-dimensional array vector value (s, d) spacing/direction value expression, in the regional area is carried out convolution with one 19 * 19 vector filter forms, these filtering forms be circular symmetric and have that (i j) is worth,-9＜=i, j＜=9.The output result of convolution be (S, image value D) and convolution kernel (I, J) absolute value after the dot product add and.This value of image maximum point be considered to inside center.The principal direction of fingerprint is to be that the modal value of the histogram of the direction (D) in the annular region of 96 pixels is estimated with radius around the center.(X Y) will be stored with principal direction P at the center of core.
In case the direction of fingerprint image and inside center (core) are positioned out, the direction of the crestal line of estimation and spacing can be rotated and displacement adjustment (step 506) according to general starting point (core) and principal direction.
In a typical fingerprint image, generally have about 3000 zones, this value can change a lot according to the size of scanning device and fingerprint.The parameter that each area inner measuring goes out can be expressed as the vector of a three-dimensional that is made of the direction of estimating the crestal line that, spacing and phase place, and these three parameters are that the mathematical modeling to fingerprint pattern obtains in 503,504 steps.The summation of all dimensions of the figure vector of the All Ranges of the fingerprint that each is complete is approximately hundreds of or thousands of, and each fingerprint is by one group of zone vector representation.Each Vector Groups to each fingerprint repeats this mapping process, and whole group storage fingerprint can be regenerated in the corresponding vector space of a multidimensional coordinate system.
No matter matching process (not appearing at Fig. 5) is checking or identification, all is the contrast (referring to preamble) between fingerprint from essence.Under the environment of vector space, the contrast between fingerprint need be measured this distance of organizing the vector between each fingerprint.This relates in the vector space of high dimension and calculates, and this needs sizable system-computed ability, can cause the expensive and inefficient of system.
An object of the present invention is, it effectively is reduced to the demand of calculating in the normal calculated performance.One aspect of the present invention is to use the dimension that multiple technologies reduce feature space, Vector Groups is projected onto in the new coordinate system, the topmost difference of data for projection is positioned at first coordinate axis (being called first pivot), and the second main difference is positioned on second coordinate axis, etc.
One aspect of the present invention is that application pivot analysis technology (PCA) is reduced to an easy-to-handle rank with the dimension in fingerprint vector space, makes the analysis to fingerprint become easy.Specifically, the aspect of model vector after each calibration is averaged to obtain a mean value model vector, M=E[V].The covariance of model vector is C=E[(V-M) (V-M) ^T], this value is the expectation of apposition of the difference V-M of model vector and mean vector, ^T is transpose of a matrix, E[] and be the expectation operational symbol.Then obtain one group of new main characteristic direction V ' by PCA.The eigenwert fabric of the proper vector that is produced by PCA has become the base group in characteristic feature space then.
Proper vector V ' can come it is compressed by getting the subclass of only including all variable part number percent K, for example the K value reasonably is set to 95% of total variable.V '=(E1, E2, E3, E4...EK) (a1, a2, a3, a4 ... aK) ^T, here a1 is one group of scalar to aK, E1 is the unit length vector (^T is transpose of a matrix as preamble) of covariance matrix to EK.K is selected come out in step 507, and total like this variable number will be less than certain number percent, and for example 95%.Scalar a1 is to calculate by the method for a standard to aK, and this method is with above-mentioned vector of unit length projection or V to be measured, for example for j proper vector of j scalar sum, aj=V^T Ej.
Another implementation of this method is that under the no supervision pattern, the coupling of system's meeting application success is learnt, and the parameter of updated stored template (its average and covariance in the characteristic feature space).The data that systematic learning arrives are more many, the ability that fingerprint is distinguished by system is more strong: the classification between similar and inhomogeneity can make its effect better by using sorter to greatest extent, sorter can be linear discriminant analysis (LDA) or non-linear center learning method, for example center LDA.
The vector of the representative fingerprint image after the calibration can be projected onto among the vector space V, then uses above-mentioned dimension and subdues technology or other substitute technologies, projects the general features SPACE V ' on, this appearance for fingerprint is constant.
After handling all images under offline mode or the online mode, every width of cloth image will be used the point of subduing on the feature space of back and show.According to the dimension of subduing, the different place between point can become obviously, thus the similarity between enhancing or inhibition point.
An advantage of the invention is, by will each area measure to fingerprint parameter project on the fingerprint space that includes those representation parameters (for example vector), can carry out flexible processing to data.Strengthening data by said method (for example dimensionality reduction) can allow the user with than the more simple method reinforcement of technology before or dwindle similarity between fingerprint.
By according to the point in the position of dimensionality reduction rear space these points are divided into groups, the propinquity between point can be used as the representative of similarity degree between parameter.Can use multiple sorting technique afterwards (step 509) cut apart in the fingerprint space.A method is the k-means clustering technique, and this is a program in two steps: each template at first is associated with or is labeled as the most close prototype in the initial set of M cluster prototype; By moving to current mark the position of these templates is upgraded then.Another iteration method is distance between applying template and the relative expansion of current cluster.Another kind method is to be grouped in the increasing cluster by layering to obtain cluster.In certain methods, the quantity M of cluster can be imported in the algorithm, perhaps is detected, just as in some hierarchical cluster analysis methods.
The reader may recognize that the present invention is good at managing data group large-scale, high complexity.By will the data relevant with the zone of fingerprint image being mapped in the typical feature space (for example representative higher-dimension coordinate system), handle these data and will become more convenient, and the difference of finding out between image also can become more easy.The influence of eliminating noise and cut in typical feature space is relative also comparatively easy.
It is that it is to the not dependence of details and pore information that the present invention also has a benefit, though concerning the accuracy that obtains fingerprint measurement or coupling, all be vital in these information system formerly---the present invention can take all factors into consideration the data relevant with pore and details really, and is irrelevant in its minimum form and these information.
The constituent (step) in these structure characteristic feature spaces can be replaced with other technology generation.The inventor is considering the method that these substitute.
Survey pore information
Position, shape and distribution that WO2005/022446 proposes pore have the uniqueness of height, and these pore information may be used solely to or carry out fingerprint recognition or checking as the side information of minutia.A kind of implementation according to an aspect of the present invention, Fig. 6 has shown the steps flow chart of surveying pore information.The step (step 502,503,504) that proposes before using, the crestal line figure can be by certain periodically clear and definite showing of mathematical model.The mode of reconstruction that therefore, can be by deducting the crestal line pattern from raw data removes (step 601) with crestal line information from original image.The information that stays will comprise pore and other background characteristics noises.General pore profile has higher density for circular (water droplet shape) and than background, yet their size and dimension can be different, and the border can be very irregular sometimes.Therefore, pore Feature Recognition consistent and strong robustness becomes a challenge.
The implementation of a first-selection of the present invention is to use the Hermite polynomial expression pore feature is carried out modeling.At first the pixel that has a high intensity values by the location is determined the candidate's or possible zone, around these pixels, put forms and with this zone marker for the candidate's or possible zone (step 602).In next step of this implementation, the application parameter model is represented the possible pore (step 603) in each zone.In the implementation of the present invention, the raw data in the zone is at first carried out modeling by two-dimentional gaussian intensity profile, and here, the maximum likelihood method that parameters such as the average of strength model and covariance can using iterative is estimated out.Another implementation is to utilize the Minimum Mean Square Error technology to estimate parameter.These two kinds of technology are for all being understandable concerning the people who is skilled in technique.Gaussian distribution can be simulated prototype feature very accurately, yet it is not enough to simulate erose feature.The Hermite polynomial expression is the accuracy of strengthening the flexibility ratio of model and then strengthening simulation.In the implementation of a first-selection, only adopt the several Hermite coefficients that begin most to come pore information is simulated.
Ideally, we can be applied in all features of simulating of determining in the step 603 and carry out fingerprint matching, yet the dust or the oil stain that have comprised on swept noise, scanner surface or the skin all can produce the feature that is similar to pore at interior many ground unrests.In order to strengthen the consistance that pore is surveyed, an implementation of the present invention is to adopt filter step (step 604).
System carries out filtering based on the shape in hole to possible pore.The implementation of a first-selection is to use the method for pivot analysis the covariance of the Hermite polynomial parameters of estimation in the simulation of step 603 pore is analyzed.The main pattern of having included the part (for example 90% or 95%) of all variablees is used for some hole filterings are gone out, and these holes are outer holes of domain transformation that the projection of those their Hermite multinomial coefficients is positioned at the feature that is selected (shape) space that is calculated by PCA.Can control shape filtering result's strict degree with a user-defined threshold value.This special implementation is a linear method of a unlikely shape difference of elimination, but the present invention does not get rid of the non-linear shape modeling technique that uses as core PCA technology.
In the step of a secondary filtering, the position of pore is considered to and marking crestal line pattern relevant (step 503).At first, only can be taken into account with the overlapping possible hole of crestal line.For a kind of implementation of first-selection, those nominal areas and crestal line overlapping Kong Caihui to a great extent are allowed to by mailing to subordinate's wave filter; Those do not have a lot of overlapping holes to be cut.Second level position filtering device considers that those more or less are positioned at the hole of crestal line center line (along the direction of crestal line).On the direction perpendicular to crestal line, the change of little position allows.With the amplitude of being permitted be proportional with the crestal line spacing.This is a predetermined parameter.What the next stage position filtering was considered is the frequency that occurs along whole piece crestal line pore, and uses and determine basic period frequency along the harmonic expansion of crestal line pore position.Pore position on the higher hamonic wave is deleted.Fundamental frequency can be by obtaining in similar, the high-quality marking or the current marking.In the implementation of first-selection, can be used for carrying out frequency analysis along the Fourier analysis of the one-dimensional signal of the position in the hole of camber line.
In the implementation of first-selection, the wave filter of these pore shapes and pore position is used according to said sequence, also can design some other sequential combination.Whether in the implementation of first-selection, all shapes and position filtering device can be used to, but all use the pore that depends on the vacation that obtains the quality of fingerprint and detect in step 601 and 603 to account for genuine ratio it.
Image calibration
Image calibration has been removed the geometric deformation on the fingerprint, and these deformation are to cause owing to obtaining image with the diverse location of fingerprint instrument from different perspectives.As mentioned before, system before calibrates fingerprint image based on detailed information, and this inevitably can fall into the problem that is caused by two X factors (for example consistance of the conversion between the details group and minutiae point).
The crestal line parameter of mentioning before image calibration of the present invention is based on, so this is irrelevant with the distribution of details.Except global change, calibration steps hereinafter described can be eliminated the elastic deformation that scanner application of force inequality, distortion and various other factors is caused by scanner distortion, fingerprint equally.In the technology based on details calibration before, these deformation and distortion are difficult to adjust with systematic method, and it is inaccurate that this can cause identifying.
Fig. 7 is the process flow diagram of the implementation of image calibrating method of the present invention.According to implementation of the present invention, the average of the expression information that two inputs of Fig. 7 are query image and its affiliated class.According to implementation of the present invention, two one of inputs of Fig. 7 are the expression information of query image or candidate image, and another is the expression information of the image template of typing before.The implementation step of back is the first-selected implementation of a kind of 1:1 coupling (checking).In these implementations, the expression information of image is a multidimensional feature group { spacing, direction, phase place } that generates with step 501-506.
In step 701, candidate template and storing template can be come to produce some zones for every width of cloth image by lattice.The feature group of each input can be divided in the group based on its relative position to nucleus.Each group will represent the image in a zone.Therefore, Qu Yu size is to the distance of core and fixed according to it.
The target of step 702 is that estimation is to the conversion parameter { convergent-divergent, rotation, translation } of each area applications, in order to compensate above-mentioned deformation and distortion processing.A kind of implementation of first-selection according to the present invention, recursive filtering technology are used to carry out transformation parameter estimation, and this conversion is the conversion that each zone is applied.Candidate image must template associated therewith with, that represent the image class that provides compare.The expression point in each zone of candidate template need with the same area of image template in corresponding expression information compare, the difference between them is indicating and eliminating the needed conversion of distortion in the specific region.
The parameter in assessing of conversion process is the process of an iteration, and each iteration of this process has all comprised prediction and upgraded two steps.Estimation before prediction steps is based in iteration upgrades being based on the error of prediction in new one-shot measurement.When before with current iteration in the difference estimated enough little in, above-mentioned steps may be considered to be tending towards identical.By above-mentioned step, just can obtain optimum transformation parameter, can calibrate the respective regions of query image and template image best by this parameter.
Estimate the parameter that and be applied to subsequently on the respective regions of query image, this zone is calibrated to respective regions (step 703) on the template.In next step (step 704), by the similarity degree of respective regions of the query image after contrast mould's image and the calibration, can calculate a calibration mark to each zone.If this calibration mark is not high enough, that just means still have some elastic deformations in the zone, therefore need the regional of this Region Segmentation Cheng Gengxiao and repeat 702 to go on foot for 704 steps, the enough height of calibration mark or data up to each zone can not be used for carrying out the recursive calculation in 702 steps very little.In case all subgroups (zone) is processed intact, just can obtain the overall situation and elastically-deformable transformation parameter by each regional transformation parameter being carried out interpolation (step 707).Then the transformation parameter after these interpolations is applied in the query image, according to using it is calibrated to average or the single template of class.
The advantage of calibration system of the present invention is, compared to previous system, it can more easily compensate because the distortion of the fingerprint image that excessive or unbalanced pressure and finger roll cause during fingerprint extraction.Lattice step subsequently makes system have higher accuracy when the finger-print region that is subjected to distortion effect is simulated, and this step is to the distance of finger print core the zone further to be divided into subregion according to the zone.The improvement of calibration process has improved the normalization effect of data groups, makes the probability that occurs by mistake mating as two width of cloth images to identical finger drop to minimum simultaneously.
The fingerprint typing
Fingerprint recognition system contains two kinds of main operational modes: typing and coupling.In the typing stage, the fingerprint of acquisition is stored in the template database, has only those features with fingerprint of the property distinguished to be extracted out here and shows with certain form.Fig. 8 is the process flow diagram of the first-selected implementation of typing pattern of the present invention.
The typing process of an implementation of the present invention starts from obtaining fingerprint image (step 801) from scanning device.In following step (step 802), image is divided into one group of zone (step 502), and according to direction and spacing that the crestal line that is estimated in method (step 503) and each zone of description before each regional crestal line pattern is simulated.Use and the same method of in step 505, describing position and the direction of location core.Then parameter group { spacing, direction } is projected in the characteristic feature space described in Fig. 5.
In an implementation of invention, if the position of projected pattern is very approaching or overlap on pattern on the template of typing before, with regard to execution in step 809 and state the typing of a repetition, for example fingerprint is entered in the database.Otherwise, uses new candidate image feature space is upgraded, and the information on classification and the same needs application of cluster step (step 509, the 510) new images is calculated again.In following step (step 806), can from the image of typing, be extracted as the detailed information such as position, direction and crestal line number of details.These information need be calibrated to the characteristic feature space equally.Step 807 is optional steps, and it is used for extracting pore information from input picture.These information comprise the features such as position, shape and distribution of pore.A concrete implementation is presented hereinbefore and is presented among Fig. 6.If the details number that detects and pore number are lower than one by user configured predetermined value, system re-executes step 811 and states a typing failure so.Otherwise, the position of image in feature space, the class under it and the information of subclass, the crestal line parameter, the position of its core and direction, details and pore information can be encoded and be stored as a template image.In order to keep the collaborative work of template and other programs, template can generate with a kind of form of multilayer, and other method or computer program just can call the Template Information as parts such as position, type and/or details directions like this.
Fingerprint matching
The fingerprint matching process has comprised the one or more fingerprints in query fingerprints and the one group of template has been compared.Mainly be divided into two patterns in handling in this section: Validation Mode (for example man-to-man matching inquiry) and recognition mode (for example matching inquiry of one-to-many).Following explanation is based on recognition mode, wants to be applied to Validation Mode, only needs do very little change to flow process.Fig. 9 is the process flow diagram of a first-selected implementation of invention.
After scanning device has obtained query image, it is decomposed into the zone of one group of part, and each zone is simulated and definite crestal line information, and then it is plotted in the characteristic feature space (step 902).On the step stated all used same method, this method is explained in the typing process.In step 903, based on the position of query image in feature space, calculate the possibility that this image belongs to certain class or subclass.The possibility that the query fingerprints Q that provides belongs to certain specific class c can use probability P (c|V (Q), M_c, C_c) expression, here M_c is the average of c class pattern, C_c is the covariance of c class pattern.V (Q) is the projection (step 508) of Q on the characteristic feature SPACE V.In one implementation, this probable value can obtain under the situation of the expansion of not considering each pattern class (for example its covariance C_c).This probability can obtain based on Euclidean distance simply so, for example P (c)=Exp[-||V (Q)-M_c||], EXP[] be the exponentiation function, || || be Euclid's vector norm or distance.Otherwise, under the situation of the expansion of considering each class, can suppose for example general a distribution or the probability model of Gaussian distribution.Following probable value is calculated by the gaussian probability distribution function, and Exp[-(V (Q)-M_c) ^T C_c^-1 (V (Q)-M_c)], here T is the matrix transpose symbol, C_c^-1 is the inverse matrix of covariance matrix C_c.This formula should be very familiar for the people of pattern classification on top of.The implementation of the reality of step 509 can advise calculating with additive method the probability of ownership or classification: P, and (c|V (Q), H_c), here H_c is distributed in parameters of probability set different under the C class pattern.No matter selection parameter or non-parametric probability model all needs with the cutting procedure of using in 510 steps relevant.This variation is under the consideration of invention intention.
If there is not satisfactory class, system is with execution in step 914 and state non-matching.Otherwise system will carry out step 905.In this step, for each satisfactory class, the crestal line parameter set credit union of query image and the set of the Mean Parameters of each class compare.For linearity and the non-linear deformation between these two parameters, the method that system's meeting application image calibration steps is described removes it.Be applied in the query image estimating the transformation parameter that then.
In step 906, from the query image after the calibration, extract detailed information, and the corresponding information in the storing template in this information and the subregion class is compared.If the quantity of detailed information very little, algorithm can be carried out step 908 so, and this step is extracted (having carried out detailed explanation in to the description of Fig. 6) to pore information.If in step 906, obtained abundant detailed information, then inquiry and storing template are compared, and in step 909, rely on detailed information and probability to generate the mark of a coupling merely.Otherwise application pore information compares the corresponding information of each template in query image and the class.
Then, system can combine probability, detailed information and/or the pore information that 903 steps calculated, and generates a comprehensive coupling mark.In a kind of implementation of first-selection, these two marks: an expression belongs to the probability of certain close pattern class c; The similarity of details/pore information of each template R_c is united in the mode that it is expressed as probability in another expression query fingerprints Q and the c class.So, P (Q matches R_c)=P (c|V (Q), H_c) x P (the details mark of Q and R_c).First probability is identical with the mark of estimation in 903 steps, and many known methods are arranged now can be probability with the similarity fraction representation based on details and/or pore position easily.Be higher than predetermined threshold as certain the coupling mark in the fruit, then system will state the coupling between query image and corresponding template.Otherwise, return step 905, and the template in the satisfactory class of the next one handled.In the implementation of a first-selection, next the order of the satisfactory class of Chu Liing (with and the template that comprises) be to be ranked according to " affiliated " mark that in step 903, calculates.All templates and query image in all satisfactory classes are finished contrast, and without any a coupling mark when being higher than predetermined threshold value, system will state one non-matching.
It should be noted that this explanation only is the principle example of an invention, it should not become restriction of the present invention, and it is purpose to explain and to understand invention.For those people to being skilled in technique, from disclosed information, can learn easily, employing also can realize corresponding function in implementation and the method for other elements of this introduction, as long as these methods are without prejudice to the principle of the invention of here introducing, invention here is defined with claim.
Claims (15)
1. method of handling fingerprint image has comprised following steps:
For each image to be processed
-by conversion and/or rotational alignment image
-image is divided into one group of zone,
For each zone of image, at least one parameter in the parameter below measuring:
The direction of-backbone line
-average ridge distance between centers of tracks
-phase place,
Store above-mentioned measured value,
For the image after all processing,
-each above-mentioned regional measured value is projected in first coordinate system of a multidimensional and in above-mentioned first coordinate system presentation video, here, the representative distance between the expression information of the relevant parameter of two width of cloth images shows the dissimilarity of respective image.
2. according to the process of claim 1 wherein that calibration steps comprises:
-biological center and the biological axle of determining fingerprint in the image
-a general reference point and general axis of reference be set
-converted image is repositioned onto the biological center of fingerprint on the general reference point
-image rotating makes the biological direction of principal axis of fingerprint consistent with the direction of axis of reference.
3. calibration steps according to claim 2, if the biological center of query fingerprints does not appear in the image, then the extrapolation of combining image part fingerprint ridge line and known fingerprint pattern are estimated the position of biological center outside image.
4. according to the described method of claim before, its feature has also comprised uses the image that one-period wave function model comes fingerprint in the simulated domain, and here, above-mentioned parameter is measured acquisition in above-mentioned model image and/or true picture.
5. fingerprint image simulation steps according to claim 4, the periodic wave function model is sinusoidal form.
6. method according to claim 5, its feature have also comprised the step that estimation error is measured, here estimation error be defined on the model with non-modeled image in different between the parameter measured.
7. according to the described method of claim 5, if the estimation error of specific region has been surpassed certain predetermined first threshold, then this area applications segmentation step second time is created this regional subregion, and in subregion, use measuring process.
8. according to the described method of claim before, the expression step is included in previously mentioned first coordinate system with vectorial V representative image, and vectorial V is corresponding with each regional measured value of every width of cloth image, and this first coordinate system has constituted vector space.
9. according to the described method of claim before, its feature has also comprised and has strengthened previously mentioned representative apart from the method for observability.
10. according to the method for the representative of the enhancing described in the claim 9 apart from observability, this step has comprised in data projection to one second coordinate system that will measure, here, two width of cloth images represent that in second coordinate system representative distance of information is bigger in first coordinate system than it.
11. according to the method described in the claim 9, its feature also comprises dimension reduction step,, subdues a dimension at least here from first coordinate system.
12. according to the reduction of the dimension described in the claim 11 step, at least one dimension is cut out by pivot analysis method (PCA).
13. according to the pivot analysis method (PCA) described in the claim 12, covariance C application characteristic vector/eigenwert is decomposed C=E[(V-M) (V-M) ^T], come structural attitude vector V ' and structural attitude SPACE V ', here V is the pattern vector, M is average pattern vector, M=E[V].
14. according to the described step of claim 9 to 13 and method, its feature has also comprised diversity factor allocation scores step, here, after each figure image intensifying step, can strengthen discrepancy score of system assignment for each according to the distance of the representative between expression information, discrepancy score is signifying the difference degree between expression information.
15. according to the method described in the claim 14, its feature has also comprised dimension and has got rid of step, here, a part of dimension of coordinate system will be not processed, have only the Partial K of non-removing property to be allowed to handle in all dimensions, the dimension that is excluded is that those discrepancy scores that cause after with it eliminating are less than the dimension of second a predetermined threshold value.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN2012100535394A CN103294987A (en) | 2012-03-05 | 2012-03-05 | Fingerprint matching method and fingerprint matching implementation mode |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN2012100535394A CN103294987A (en) | 2012-03-05 | 2012-03-05 | Fingerprint matching method and fingerprint matching implementation mode |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN103294987A true CN103294987A (en) | 2013-09-11 |
Family
ID=49095826
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN2012100535394A Pending CN103294987A (en) | 2012-03-05 | 2012-03-05 | Fingerprint matching method and fingerprint matching implementation mode |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN103294987A (en) |
Cited By (17)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN105426877A (en) * | 2015-12-22 | 2016-03-23 | 金虎林 | Method and system for information recognition applying sweat gland position information |
| CN105740821A (en) * | 2016-01-29 | 2016-07-06 | 广州立为信息技术服务有限公司 | Fingerprint identification method and system |
| CN106104574A (en) * | 2016-02-25 | 2016-11-09 | 深圳市汇顶科技股份有限公司 | Fingerprint identification method, device and terminal |
| CN106250890A (en) * | 2016-09-23 | 2016-12-21 | 南昌欧菲生物识别技术有限公司 | A kind of fingerprint identification method and device |
| CN107066971A (en) * | 2017-04-17 | 2017-08-18 | 惠州Tcl移动通信有限公司 | It is a kind of that fingerprint detection control method and system are singly referred to based on mobile terminal |
| CN107743130A (en) * | 2013-11-06 | 2018-02-27 | 阿里巴巴集团控股有限公司 | The method, apparatus and system of a kind of fingerprint matching |
| WO2018099077A1 (en) * | 2016-12-01 | 2018-06-07 | 京东方科技集团股份有限公司 | Image matching method, device and system, and storage medium |
| CN108667768A (en) * | 2017-03-29 | 2018-10-16 | 腾讯科技（深圳）有限公司 | A kind of recognition methods of network application fingerprint and device |
| CN109074493A (en) * | 2017-03-10 | 2018-12-21 | 指纹卡有限公司 | Inhibit the damage data in fingerprint image |
| CN109101889A (en) * | 2018-07-12 | 2018-12-28 | 苏海英 | Finger scan mechanism based on dust analysis |
| CN110598666A (en) * | 2019-09-19 | 2019-12-20 | 哈尔滨工业大学（深圳） | Method, device and system for matching high-resolution fingerprint sweat pores and storage medium |
| CN111318697A (en) * | 2018-12-13 | 2020-06-23 | 通用电气公司 | Method for monitoring a molten bath using fractal dimension |
| CN109151193B (en) * | 2018-08-13 | 2020-09-22 | 维沃移动通信有限公司 | A kind of alarm clock control method and mobile terminal |
| CN112307880A (en) * | 2019-08-01 | 2021-02-02 | 联咏科技股份有限公司 | Electronic circuit, electronic device and method for sensing at least one fingerprint image |
| CN113537178A (en) * | 2021-09-16 | 2021-10-22 | 南通市海鸥救生防护用品有限公司 | Face picture compensation identification method based on ship security data identification |
| TWI754241B (en) * | 2020-02-27 | 2022-02-01 | 大陸商敦泰電子(深圳)有限公司 | A method, a device for extracting features of fingerprint images and computer-readable storage medium |
| US12148237B2 (en) | 2019-08-01 | 2024-11-19 | Novatek Microelectronics Corp. | Imaging control circuit for collecting object image data for an object |
Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20020041700A1 (en) * | 1996-09-09 | 2002-04-11 | Therbaud Lawrence R. | Systems and methods with identity verification by comparison & interpretation of skin patterns such as fingerprints |
| US20040170306A1 (en) * | 2003-02-28 | 2004-09-02 | Nec Corporation | Fingerprint verification device and method for the same |
| CN101114335A (en) * | 2007-07-19 | 2008-01-30 | 南京大学 | All-angle fast fingerprint recognition method |
| CN101276411A (en) * | 2008-05-12 | 2008-10-01 | 北京理工大学 | Fingerprint identification method |
| CN101329727A (en) * | 2008-06-27 | 2008-12-24 | 哈尔滨工业大学 | Fingerprint Recognition Method Combining Points and Lines |
| GB2450479A (en) * | 2007-06-22 | 2008-12-31 | Warwick Warp Ltd | Fingerprint recognition including preprocessing an image by justification and segmentation before plotting ridge characteristics in feature space |
-
2012
- 2012-03-05 CN CN2012100535394A patent/CN103294987A/en active Pending
Patent Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20020041700A1 (en) * | 1996-09-09 | 2002-04-11 | Therbaud Lawrence R. | Systems and methods with identity verification by comparison & interpretation of skin patterns such as fingerprints |
| US20040170306A1 (en) * | 2003-02-28 | 2004-09-02 | Nec Corporation | Fingerprint verification device and method for the same |
| GB2450479A (en) * | 2007-06-22 | 2008-12-31 | Warwick Warp Ltd | Fingerprint recognition including preprocessing an image by justification and segmentation before plotting ridge characteristics in feature space |
| CN101114335A (en) * | 2007-07-19 | 2008-01-30 | 南京大学 | All-angle fast fingerprint recognition method |
| CN101276411A (en) * | 2008-05-12 | 2008-10-01 | 北京理工大学 | Fingerprint identification method |
| CN101329727A (en) * | 2008-06-27 | 2008-12-24 | 哈尔滨工业大学 | Fingerprint Recognition Method Combining Points and Lines |
Cited By (31)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN107743130A (en) * | 2013-11-06 | 2018-02-27 | 阿里巴巴集团控股有限公司 | The method, apparatus and system of a kind of fingerprint matching |
| CN107743130B (en) * | 2013-11-06 | 2020-04-07 | 阿里巴巴集团控股有限公司 | Fingerprint matching method, device and system |
| CN105426877A (en) * | 2015-12-22 | 2016-03-23 | 金虎林 | Method and system for information recognition applying sweat gland position information |
| CN105426877B (en) * | 2015-12-22 | 2019-09-10 | 金虎林 | Utilize the information identifying method and system of sweat gland location information |
| CN105740821A (en) * | 2016-01-29 | 2016-07-06 | 广州立为信息技术服务有限公司 | Fingerprint identification method and system |
| WO2017143571A1 (en) * | 2016-02-25 | 2017-08-31 | 深圳市汇顶科技股份有限公司 | Fingerprint identification method, device, and terminal |
| CN106104574B (en) * | 2016-02-25 | 2018-06-12 | 深圳市汇顶科技股份有限公司 | Fingerprint identification method, device and terminal |
| CN106104574A (en) * | 2016-02-25 | 2016-11-09 | 深圳市汇顶科技股份有限公司 | Fingerprint identification method, device and terminal |
| CN106250890A (en) * | 2016-09-23 | 2016-12-21 | 南昌欧菲生物识别技术有限公司 | A kind of fingerprint identification method and device |
| CN106250890B (en) * | 2016-09-23 | 2020-05-05 | 南昌欧菲生物识别技术有限公司 | Fingerprint identification method and device |
| WO2018099077A1 (en) * | 2016-12-01 | 2018-06-07 | 京东方科技集团股份有限公司 | Image matching method, device and system, and storage medium |
| US10679364B2 (en) | 2016-12-01 | 2020-06-09 | Boe Technology Group Co., Ltd. | Image matching method, image matching apparatus, image matching system, and storage medium |
| CN109074493A (en) * | 2017-03-10 | 2018-12-21 | 指纹卡有限公司 | Inhibit the damage data in fingerprint image |
| CN108667768A (en) * | 2017-03-29 | 2018-10-16 | 腾讯科技（深圳）有限公司 | A kind of recognition methods of network application fingerprint and device |
| CN108667768B (en) * | 2017-03-29 | 2022-04-29 | 腾讯科技（深圳）有限公司 | Network application fingerprint identification method and device |
| CN107066971A (en) * | 2017-04-17 | 2017-08-18 | 惠州Tcl移动通信有限公司 | It is a kind of that fingerprint detection control method and system are singly referred to based on mobile terminal |
| CN107066971B (en) * | 2017-04-17 | 2020-07-14 | 惠州Tcl移动通信有限公司 | Single-finger fingerprint inspection control method and system based on mobile terminal |
| CN109101889B (en) * | 2018-07-12 | 2019-08-02 | 新昌县哈坎机械配件厂 | Finger scan mechanism based on dust analysis |
| CN109101889A (en) * | 2018-07-12 | 2018-12-28 | 苏海英 | Finger scan mechanism based on dust analysis |
| CN109151193B (en) * | 2018-08-13 | 2020-09-22 | 维沃移动通信有限公司 | A kind of alarm clock control method and mobile terminal |
| CN111318697B (en) * | 2018-12-13 | 2023-01-03 | 通用电气公司 | Method for monitoring a weld pool using fractal dimension |
| CN111318697A (en) * | 2018-12-13 | 2020-06-23 | 通用电气公司 | Method for monitoring a molten bath using fractal dimension |
| CN112307880A (en) * | 2019-08-01 | 2021-02-02 | 联咏科技股份有限公司 | Electronic circuit, electronic device and method for sensing at least one fingerprint image |
| TWI826061B (en) * | 2019-08-01 | 2023-12-11 | 聯詠科技股份有限公司 | Electronic circuit, an electronic device and a method for sensing a fingerprint image from a panel |
| US11854451B2 (en) | 2019-08-01 | 2023-12-26 | Novatek Microelectronics Corp. | Electronic circuit having display driving function, touch sensing function and fingerprint sensing function |
| US12008939B2 (en) | 2019-08-01 | 2024-06-11 | Novatek Microelectronics Corp. | Electronic circuit having display driving function, touch sensing function and fingerprint sensing function |
| US12148237B2 (en) | 2019-08-01 | 2024-11-19 | Novatek Microelectronics Corp. | Imaging control circuit for collecting object image data for an object |
| CN110598666B (en) * | 2019-09-19 | 2022-05-10 | 哈尔滨工业大学（深圳） | Method, device and system for matching high-resolution fingerprint sweat pores and storage medium |
| CN110598666A (en) * | 2019-09-19 | 2019-12-20 | 哈尔滨工业大学（深圳） | Method, device and system for matching high-resolution fingerprint sweat pores and storage medium |
| TWI754241B (en) * | 2020-02-27 | 2022-02-01 | 大陸商敦泰電子(深圳)有限公司 | A method, a device for extracting features of fingerprint images and computer-readable storage medium |
| CN113537178A (en) * | 2021-09-16 | 2021-10-22 | 南通市海鸥救生防护用品有限公司 | Face picture compensation identification method based on ship security data identification |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN103294987A (en) | Fingerprint matching method and fingerprint matching implementation mode | |
| CN101980250B (en) | Method for identifying target based on dimension reduction local feature descriptor and hidden conditional random field | |
| US11113505B2 (en) | Palm print image matching techniques | |
| Dibeklioglu et al. | 3D facial landmarking under expression, pose, and occlusion variations | |
| Maltoni et al. | Handbook of fingerprint recognition | |
| Li et al. | Multiscale Features for Approximate Alignment of Point-based Surfaces. | |
| EP2174261B1 (en) | Fingerprint matching method and apparatus | |
| US8280150B2 (en) | Method and apparatus for determining similarity between surfaces | |
| Lategahn et al. | Texture classification by modeling joint distributions of local patterns with Gaussian mixtures | |
| Jung et al. | Noisy and incomplete fingerprint classification using local ridge distribution models | |
| CN108830888B (en) | Coarse matching method based on improved multi-scale covariance matrix characteristic descriptor | |
| JPH05159065A (en) | Method and system for obtaining and recognizing image shape | |
| JPH10177650A (en) | Device for extracting picture characteristic, device for analyzing picture characteristic, and system for collating picture | |
| JP2008310814A (en) | Fingerprint verification system and fingerprint verification method | |
| JP3914864B2 (en) | Pattern recognition apparatus and method | |
| CN117571341B (en) | System and method for detecting omnibearing wear of tire | |
| CN118823126B (en) | A method for positioning PCB solder-free optical devices | |
| Giachetti | Effective characterization of relief patterns | |
| CN117541807A (en) | Terrain characteristic line extraction method based on active learning | |
| CN107784284B (en) | Face recognition method and system | |
| Wang et al. | Joint head pose and facial landmark regression from depth images | |
| JP2003216931A (en) | Specific pattern recognizing method, specific pattern recognizing program, specific pattern recognizing program storage medium and specific pattern recognizing device | |
| Srivastava et al. | Drought stress classification using 3D plant models | |
| CN119204846A (en) | Prefabricated building quality monitoring system based on BIM and point cloud integration | |
| CN103064857A (en) | Image query method and image query equipment |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| C06 | Publication | ||
| PB01 | Publication | ||
| C10 | Entry into substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| C02 | Deemed withdrawal of patent application after publication (patent law 2001) | ||
| WD01 | Invention patent application deemed withdrawn after publication |
Application publication date: 20130911 |