CN109741273A - A kind of mobile phone photograph low-quality images automatically process and methods of marking - Google Patents
A kind of mobile phone photograph low-quality images automatically process and methods of marking Download PDFInfo
- Publication number
- CN109741273A CN109741273A CN201811596737.9A CN201811596737A CN109741273A CN 109741273 A CN109741273 A CN 109741273A CN 201811596737 A CN201811596737 A CN 201811596737A CN 109741273 A CN109741273 A CN 109741273A
- Authority
- CN
- China
- Prior art keywords
- image
- quality images
- low
- answer
- marking
- 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 claims abstract description 55
- 230000011218 segmentation Effects 0.000 claims abstract description 14
- 238000012360 testing method Methods 0.000 claims abstract description 10
- 238000012545 processing Methods 0.000 claims description 25
- 238000011084 recovery Methods 0.000 claims description 9
- 238000012937 correction Methods 0.000 claims description 5
- 239000011159 matrix material Substances 0.000 claims description 4
- 238000000926 separation method Methods 0.000 claims description 4
- 238000003709 image segmentation Methods 0.000 claims description 3
- 238000013461 design Methods 0.000 abstract description 2
- 235000013350 formula milk Nutrition 0.000 description 30
- 230000000717 retained effect Effects 0.000 description 5
- 230000009466 transformation Effects 0.000 description 4
- 238000005286 illumination Methods 0.000 description 3
- 230000000694 effects Effects 0.000 description 2
- 241001270131 Agaricus moelleri Species 0.000 description 1
- 230000003044 adaptive effect Effects 0.000 description 1
- 238000004364 calculation method Methods 0.000 description 1
- 239000011248 coating agent Substances 0.000 description 1
- 238000000576 coating method Methods 0.000 description 1
- 230000021615 conjugation Effects 0.000 description 1
- 238000010586 diagram Methods 0.000 description 1
- 238000010191 image analysis Methods 0.000 description 1
- 230000010365 information processing Effects 0.000 description 1
- 238000012804 iterative process Methods 0.000 description 1
- 239000004576 sand Substances 0.000 description 1
Landscapes
- Image Processing (AREA)
Abstract
The present invention provides a kind of mobile phone photograph low-quality images automatically process and methods of marking, belong to computer vision field.Test blank information and model answer are saved in the database；Answering card, which is photographed, is sent to server end, denoises to the low-quality images；Rotation and twist correcting are carried out to the image after denoising；Image after rotation and twist correcting is divided in order；Model answer in identified objective item answer and database is compared into automatic scoring.The present invention is after specifically design to blank volume format, and to taking pictures, resulting low-quality image can be automatically processed, and has shape by a relatively large margin, ratio variation that can also carry out paper segmentation and automatic scoring even if the image.It can effectively reduce the requirement to image capture device, expanded the applicable scene of automatic marking.
Description
Technical field
The invention belongs to technical field of computer information processing, more particularly to automatically processing for mobile phone photograph low-quality images
With methods of marking.
Background technique
It is the answer paper or answering card by scanner scanning student, then to card-coating during traditional examination is goed over examination papers
Topic type carry out computer automatic marking, subjective item is manually read and made comments, finally by marking system calculating total score typing achievement system
System.The method to go over files is more demanding to image capture device, it is desirable that acquired image is smooth, requires acquisition image process tight
Lattice.
Summary of the invention
The technical problem to be solved by the present invention is to provide a kind of mobile phone photograph low-quality images for the above-mentioned prior art
It automatically processes and methods of marking, answer sheet can be uploaded after student's mobile phone photograph, paper is divided in server end completion
It cuts and objective item scores, solving must be scanned in the prior art with scanner, and cost of equipment is high, not enough easily be asked
Topic.
The present invention solves the above problems used technical solution are as follows: a kind of mobile phone photograph low-quality images restore and automatic place
Reason method, described method includes following steps:
Step 1 saves test blank and model answer information in the database of server end；
Step 2, by the resulting low-quality images upload server of mobile phone photograph；
Step 3 denoises image in server end, rotational correction, deformation recovery processing；
Step 4 carries out deblurring processing to low-quality images；
Step 5 carries out accurate image segmentation to image in server end；
Model answer in identified topic answer and database is compared simultaneously automatic scoring by step 6；
Processing result is saved in server end by step 7.
Preferably, step 1 specifically comprises the following steps:
Paper region is provided with black square locating piece, has several above and below in the of every page, and size is fixed；
As needed, answer region outline border carries out overstriking, and thickness is fixed；
As needed, answering card black anchor point block is aligned with answer region outline border；
As needed, each answer unit of answering card is provided with several rows, several columns.
Preferably, step 3 specifically comprises the following steps:
Entire image processing: denoising entire image, Slant Rectify, and deformation restores.
Preferably, the entire image processing specifically comprises the following steps:
Low-quality images are denoised；
The position of black square locating piece is accurately positioned in the image denoised with integral nomography；
Deformation recovery is carried out to image according to black square locating piece position.
Preferably, it is described to low-quality images carry out denoising include the following steps:
Step 1: low-quality images are divided into several subgraphs, each subgraph is further subdivided into multiple neighborhoods, every
Otsu algorithm is used on a zonule, and each zonule is divided into background and prospect；
Step 2: using the intensity profile of former test blank as histogram specification standard, low-quality paper image is carried out
Histogram specification processing: if the histogram distribution of the window of paper image to be processed and former paper image respective window is straight
Side's figure distribution is close, then the image to the window is not dealt with, otherwise carries out histogram specification.
Preferably, described that the processing of low-quality images deblurring is specifically comprised the following steps:
Motion blur is eliminated using based on the algorithm of brightness of image and gradient priori, is enabled:
F (x)=| | x | |0
Wherein x indicates that image, f (x) indicate non-zero pixels number in image,
Using priori knowledge, have:
In formulaIndicating image gradient, σ is parameter,
F (x) is added in traditional deblurring model as regularization term, it may be assumed that
Wherein x indicates clear image, and k is fuzzy core, and λ is parameter；
It solves the model and obtains the image after deblurring, method for solving is as follows:
X, k are separated first, obtained:
And
Available x and fuzzy core k is solved for being 1. 2. iterated after above-mentioned separation, wherein each iteration is asked
Xie Shi, fixed one of another parameter of parametric solution.
Preferably, the accurate positionin black square locating piece position in the image denoised with integral nomography is specific
Include the following steps:
The integrogram of rectangular window: defining the black picture element point value in bianry image is 1, and white pixel point value is 0, rectangle
I (x, y) in the integrogram of window at any point (x, y) indicates the sum of this upper left corner all pixels in original image:
From top to bottom from given image, the integral image formula being from left to right calculated is as follows:
I (x, y)=i (x-1, y)+i (x, y)+i (x, y-1)-i (x-1, y-1)
After obtaining integrogram, the pixel of the black pixel point in image in any rectangular area and pass through following formula meter
It calculates:
Sum (m, n)=i (x, y)+i (u, v)-i (x, v)-i (u, y)
Wherein rectangle size be m=x-u, n=y-v, upper left corner of the rectangle coordinate be (u, v), bottom right angular coordinate be (x,
y)。
Preferably, described that image progress deformation recovery is included the following steps: according to tag block position
Perspective transform: the perspective transform of image is turned to formula form:
In formula, (x ', y ') indicate that the picture point coordinate after deformation, (x, y) indicate the coordinate of corresponding points in original image.
A~l is unknown parameter,
Above formula is expressed as matrix multiple:
There are eight unknown parameter a~f in the formula, m, l need to solve, and before being corrected image, need to obtain original image
Corresponding four tag block coordinates in image, eight obtained coordinates are remembered after four tag block coordinates and deformation as in
Are as follows: the coordinate points in original image: (x1, y1), (x2, y2), (x3, y3), (x4, y4), the coordinate points after deformation: (x1 ', y1 '),
(x2 ', y2 '), (x3 ', y3 '), (x4 ', y4 '), then available:
The parameter that perspective transform can be obtained is solved, the progress perspective transform of the image of deformation can be rectified after obtaining parameter
Just, available unstrained image.
Preferably, step 4 specifically comprises the following steps:
Answer region segmentation: the paper image after restoring to deformation is split according to certain rule, obtains answering for piecemeal
Inscribe region；
Divide in answer region: for being divided into several rows, the gap-filling questions of ordered series of numbers, according to the underscore at its corresponding answer,
Answer region is split, answer content is obtained.
Preferably, the answer region segmentation specifically comprises the following steps:
Every page of right boundary is estimated according to every page of rectangular locating piece；
Answer region up-and-down boundary is positioned according to the frame of overstriking up and down；
Paper image is cut according to obtained zone boundary.
Preferably, divide in the answer region and specifically comprise the following steps:
Every page of right boundary is estimated according to every page of rectangular locating piece；
Answer region up-and-down boundary is positioned according to the frame of overstriking up and down；
Divided in order according to blank coil information in database and obtained zone boundary information.
Preferably, step 5 specifically comprises the following steps:
Position application area position in image；
According to position is smeared, corresponding answer is converted to according to certain rule；
Model answer in comparison database, is scored.
Compared with the prior art, the advantages of the present invention are as follows:
The present invention is after specifically design to blank volume format, and to taking pictures, resulting low-quality image can be located automatically
Reason has shape by a relatively large margin, ratio variation that can also carry out paper segmentation and automatic scoring even if the image.It can effectively reduce
The applicable scene of automatic marking has been expanded in requirement to image capture device.
Detailed description of the invention
Fig. 1 is a kind of flow chart of mobile phone photograph paper low-quality image recovery and dividing method of the present invention；
Fig. 2 is the flow chart of a preferred embodiment of server end according to the invention.
Specific embodiment
The present invention will be described in further detail below with reference to the embodiments of the drawings.
As shown in Figure 1 and Figure 2, a kind of low-quality image based on mobile phone photograph automatically process and scoring side
Method, the system for implementing this method includes cell phone client and server end, be the described method comprises the following steps:
Test blank information and model answer are saved in the database of server end；
The resulting low-quality images upload server of mobile phone photograph, received server-side described image；
Image is denoised in server end, rotational correction, deformation restores；
Accurate image segmentation is carried out to image in server end；
By the model answer comparison in identified multiple-choice question answer and database and automatic scoring；
Processing result is saved in server end.
Test blank exists in digitized form, and model answer saves in digitized form, for commenting multiple-choice question
Timesharing comparison.Other than multiple-choice question, the image after segmentation is saved as unit of topic, and wherein multiple-choice question part is with score shape
Formula saves.
The information of the test blank includes:
Paper region is provided with black square locating piece, is provided with several above and below in the of every page, and size is fixed；
As needed, answer region outline border carries out overstriking, and thickness is fixed；
As needed, answering card black anchor point block is aligned with answer region outline border；
As needed, several rows, several columns are arranged in each answer unit of answering card；
It is described that image is denoised in server end, rotational correction, deformation recovery include:
Denoising: the noise and shade in low-quality image uploaded according to Adaptive Thresholding removal, it is subsequent fixed to facilitate
Position block identification；
Noise and shade in the low-quality image of the removal upload include the following steps:
Since condition limits, the image that student uploads can have local shades, and local overexposure etc., which influences subsequent locating piece, to be known
Other situation.Therefore the present invention needs first to remove shade to low-quality images, eliminates the pretreatment such as overexposure, and hereinafter referred to as illumination is located in advance
Reason.Illumination pretreatment needs to be divided into two steps, and the first step is to do preliminary treatment to low-quality images using local thresholding method, and second step makes
Provide that method further eliminates shade and overexposure in image with histogram.
Step 1: image is divided into several subgraphs (F1, F2 ..., Fn), each subgraph is further subdivided into more
A neighborhood (M1, M2 ..., Mn) uses Otsu algorithm on each zonule Mi (i=1...n), and Otsu algorithm step is:
If the pixel sum of piece image is S, r (i) indicates the frequency that the point of i gray scale occurs, then each gray level
The frequency occurred in piece image is:
The point in image is divided into prospect class and background classes by threshold value, it is assumed that include t point, background classes packet in prospect class
Containing S-t point, then the frequency p that foreground and background occursa(t),pb(t) formula is represented by
The gray average m of prospect class and background classesa(t),mb(t) and total mean value mTIt is respectively:
In formula:
The inter-class variance σ of background classes and prospect class2(t) it is:
σ2(t)=ma(t)(ma(t)-mT)2+mb(t)(mb(t)-mT)2=
ma(t)*mb(t)(ma(t)-mb(t))2。
Optimal segmenting threshold T ' is obtained when the maximum that inter-class variance takes, that is:
It can be obtained by multiple threshold values (t1, t2 ..., tn) using algorithm above on each region Mi, use the threshold value
The regions can be divided into background and prospect by sequence.
Step 2: by the processing of step 1, a degree of recovery is had been obtained in low-quality images, but in order to complete
The negative effect of bring may be identified to subsequent locating piece by eliminating image local shade and overexposure, it is also necessary to do image into one
The processing of step.Image histogram is a kind of important image analysis processing tool, it passes through each pixel in statistics piece image
The frequency that gray level occurs describes the gray-scale watermark of image.Histogram specification method is by providing image warp to be processed
Intensity profile of crossing that treated restores low-quality images.The algorithm basic principle is as follows:
It enables P (s) and P (t) is respectively the grey level probability density function of image to be processed and former paper image.First to be processed
Image and former paper image make histogram equalization processing.Then have:
U=F-1(U)；
Due to being all to carry out histogram equalization processing, treated original image probability density function and desired image it is general
Rate density function is equal.Thus it is possible to replace formula u=F with transformed original image gray level s-1(U) U in.
That is:
U'=F-1(s)
Gray level u ' at this time is the gray level intentionally got.For consecutive image, emphasis is to obtain inverse transformation, and right
In discrete picture, corresponding expression formula are as follows:
Z=T-1(v)。
N indicates pixel sum, P (z in formulak) indicate that the frequency that k grades of gray scales occur, grey level range are (0, l-1).
Localized region is selectively handled in the present invention: if the histogram distribution of the window of image to be processed
It is close with the histogram distribution of former paper image respective window, then the image to the window is not dealt with, otherwise need to carry out
Histogram specification.Since former paper is blank, take pictures gained image in answer region by full-filling, therefore the step only to original
It is handled in paper for non-empty white region, histogram specification is not made to the region in former paper being blank.It comprises the concrete steps that:
A rectangular scanning window is defined, sets width as odd number, size is (2d+1) * (2d+1).Consider in the window
Image mean value and variance, it is assumed that image center is (u, v), and the pixel value at point (i, j) is p (i, j), then, the window
Histogram is defined as:
Wherein ζ=0,1,2,., 256 }, S indicates sum of all pixels (2d+1) * (2d+1) in rectangular window.huv(g) reflect
The intensity profile situation of window.The histogram distribution of former test blank is obtained, this distribution is as specified straight in subsequent processing
Fang Tu.
The image mean value m of the windowuvAnd variances sigma2It is respectively as follows:
Respectively obtain the mean value and variance m of paper to be processeds,The mean value and variance m of former paper rectangular windowd,
If msAnd mdIt is close andWithIt is close, then illustrating that the window is similar with former paper, do not need to carry out histogram specification
Processing, otherwise, needs to carry out histogram specification processing.
Deformation restores: after navigating to tag block coordinate, according to the corresponding locating piece position in blank coil, according to certain rule
Carry out perspective transform；
The locative marking block coordinate specifically comprises the following steps:
The integrogram of rectangular window: defining the black picture element point value in bianry image is 1, and white pixel point value is 0.Rectangle
I (x, y) in the integrogram of window at any point (x, y) indicates the sum of this upper left corner all pixels in original image:
From top to bottom from given image, the integral image formula being from left to right calculated is as follows:
I (x, y)=i (x-1, y)+i (x, y)+i (x, y-1)-i (x-1, y-1)
After obtaining integrogram, the pixel of the black pixel point in image in any rectangular area and pass through following formula meter
It calculates:
Sum (m, n)=i (x, y)+i (u, v)-i (x, v)-i (u, y)
Wherein rectangle size be m=x-u, n=y-v, upper left corner of the rectangle coordinate be (u, v), bottom right angular coordinate be (x,
y)。
It is described to specifically comprise the following steps: according to certain rule progress perspective transform
Perspective transform: general transformation for mula in perspective transform are as follows:
In formula, (x ', y ') indicate that the picture point coordinate after deformation, (x, y) indicate the coordinate of corresponding points in original image.A~l
It is unknown parameter.
Above formula is expressed as matrix multiple:
There are eight unknown parameter a~f in the formula, m, l need to solve, and before being corrected image, need to obtain original image
Corresponding four tag block coordinates in image after four tag block coordinates and deformation as in.Eight obtained coordinates are remembered
Are as follows: the coordinate points in original image: (x1, y1), (x2, y2), (x3, y3), (x4, y4), the coordinate points after deformation: (x1 ', y1 '),
(x2 ', y2 '), (x3 ', y3 '), (x4 ', y4 '), then available:
The parameter that perspective transform can be obtained is solved, the progress perspective transform of the image of deformation can be rectified after obtaining parameter
Just, available unstrained image.
It is described to include: to the processing of low-quality images deblurring
Since condition limits, the low-quality images that student uploads can have motion blur, and the motion blur of image will lead to fixed
Position block, text are unintelligible, it is therefore desirable to eliminate the motion blur of paper image.The present invention is based on brightness of image and ladder using a kind of
The algorithm of priori is spent to eliminate motion blur.For the paper image after illumination pretreatment, paper background should be white, text
Region is black.If motion blur does not occur for text image, Luminance Distribution should have the following characteristics that its pixel intensity should have two
The characteristics of pole breaks up, most pixels are white and black.If image motion is fuzzy, then the image after fuzzy
Pixel Distribution value does not have the characteristics of sparse distribution then, has a certain number of number of pixels in multiple gray values.Text diagram
The gradient of picture then has the following characteristics that the image gradient before motion blur should be largely 0, and the text image gradient after obscuring is then
Will not have the characteristics that above.Based on the above observation, enable:
F (x)=| | x | |0
Wherein x indicates that image, f (x) indicate non-zero pixels number in image.
Priori knowledge more than utilization, has:
In formulaIndicate image gradient, σ is parameter.
F (x) is added in traditional deblurring model as regularization term, it may be assumed that
Wherein x indicates clear image, and k is fuzzy core, and λ is parameter.It solves the model and obtains the image after deblurring.It solves
Method is as follows:
X, k are separated first, obtained:
And
Available x and fuzzy core k is solved for being 1. 2. iterated after above-mentioned separation.Wherein, each iteration is asked
Xie Shi, fixed one of another parameter of parametric solution.
1. formula calculation method is as follows: utilizing half secondary split method, introduce auxiliary variable m, g, m and g correspond respectively to figure
Picture and image gradient are corresponding, obtain:
Work as parameter beta in this up-to-date style, μ level off to infinity when, 3. the solution of formula is just close to the solution of 1. formula.
3. formula separating variables are incited somebody to action, are obtained:
5. 6. thus the 3. formula for being difficult to direct solution is disassembled, respectively obtains 4., 3. direct solution is converted to repeatedly
5. 6. 4. in generation, solves, in each iterative process, fix other irrelevant variables, i.e. when solution x, fixed m, g were fixed when solving m
X, g, when solving g, fixed x, m.4. 5. 6. the solution of formula is as follows:
When beginning, m, g are initialized to null matrix, when solving x, m, g are substituted into as the known solution of iteration, using most
Small square law solves 4. formula, with analytic solutions:
Wherein F () indicates Fourier transformation, F-1() indicates inverse Fourier transform,Indicate conjugation, Respectively indicate the horizontal difference in vertical direction.
After obtaining x, using x as known conditions, then 5. 6. formula is exactly a the problem of minimizing pixel-by-pixel, m is solved, g holds
Be easy to get to:
2. the method for solving of formula is as follows: ibid, solving this formula and need to introduce auxiliary variable s, p, will 2. formula rewrite are as follows:
After introducing auxiliary variable, then by k, s, p separation, obtain:
Fixed irrelevant variable is equally taken, the method for alternating iteration solves above-mentioned three formula, and method with 1., obtains completely
The solution of formula.
The present invention only needs 20 iteration to can be obtained by preferable deblurring effect in actual use.
It is described to include: to image progress Accurate Segmentation
Answer region segmentation: the paper image after restoring to deformation is split according to certain rule, obtains answering for piecemeal
Region is inscribed, when segmentation suitably relaxes the right boundary in answer region, is effectively retained all answering informations, avoids omitting；
Divide in answer region: for being divided into several rows, the gap-filling questions of ordered series of numbers, according to the underscore at its corresponding answer,
Each topic destination locations to be split are calculated, answer region is split, answer content is obtained.
The answer region segmentation specifically comprises the following steps:
Every page of right boundary is estimated according to every page of rectangular locating piece；
Right boundary is suitably relaxed, all answering informations are effectively retained, avoids omitting；
Answer region up-and-down boundary is positioned according to the frame of overstriking up and down；
Up-and-down boundary is suitably relaxed, all answering informations are effectively retained, avoids omitting；
If the unit is final cutting unit, each answer region segmentation is completed after cutting, segmentation terminates；
Divide in the answer region and specifically comprises the following steps:
Every page of right boundary is estimated according to every page of rectangular locating piece；
Right boundary is suitably relaxed, all answering informations are effectively retained, avoids omitting；
Answer region up-and-down boundary is positioned according to the frame of overstriking up and down；
Up-and-down boundary is suitably relaxed, all answering informations are effectively retained, avoids omitting；
If the unit is not final cutting unit, it is split, is completed after cutting every according to blank coil information in database
Divide in a answer region；
The multiple-choice question answer by identification and model answer, which compare and form automatic scoring result, specifically includes following step
It is rapid:
Position application area position in image；
According to position is smeared, corresponding answer is converted to according to certain rule；
Model answer in comparison database, is scored.
A kind of low-quality image based on mobile phone photograph automatically process and methods of marking, the hardware for implementing this method are set
Standby includes paper image capture device, paper information storing device.
In any of the above-described scheme preferably, the paper image capture device includes Android mobile phone, iPhone, puts down
One of plate computer, need install present invention provide that client software.
In any of the above-described scheme preferably, the paper information storing device is for saving test blank information, mark
Quasi- answer, segmentation result.
In addition to the implementation, all to use equivalent transformation or equivalent replacement the invention also includes there is an other embodiments
The technical solution that mode is formed should all be fallen within the scope of the hereto appended claims.
Claims (12)
1. a kind of mobile phone photograph low-quality images automatically process and methods of marking, which is characterized in that the method includes walking as follows
It is rapid:
Step 1 saves test blank and model answer information in the database of server end；
Step 2, by the resulting low-quality images upload server of mobile phone photograph；
Step 3 denoises image in server end, rotational correction, deformation recovery processing；
Step 4 carries out deblurring processing to low-quality images；
Step 5 carries out accurate image segmentation to image in server end；
Model answer in identified topic answer and database is compared simultaneously automatic scoring by step 6；
Processing result is saved in server end by step 7.
2. a kind of mobile phone photograph low-quality images according to claim 1 automatically process and methods of marking, which is characterized in that
Step 1 specifically comprises the following steps:
Paper region is provided with black square locating piece, has several above and below in the of every page, and size is fixed；
As needed, answer region outline border carries out overstriking, and thickness is fixed；
As needed, answering card black anchor point block is aligned with answer region outline border；
As needed, each answer unit of answering card is provided with several rows, several columns.
3. a kind of mobile phone photograph low-quality images according to claim 1 automatically process and methods of marking, which is characterized in that
Step 3 specifically comprises the following steps:
Entire image processing: denoising entire image, Slant Rectify, and deformation restores.
4. a kind of mobile phone photograph low-quality images according to claim 3 automatically process and methods of marking, which is characterized in that
The entire image processing specifically comprises the following steps:
Low-quality images are denoised；
The position of black square locating piece is accurately positioned in the image denoised with integral nomography；
Deformation recovery is carried out to image according to black square locating piece position.
5. a kind of mobile phone photograph low-quality images according to claim 4 automatically process and methods of marking, which is characterized in that
It is described to low-quality images carry out denoising include the following steps:
Low-quality images: being divided into several subgraphs by step 1, each subgraph is further subdivided into multiple neighborhoods, each small
Otsu algorithm is used on region, and each zonule is divided into background and prospect；
Step 2: using the intensity profile of former test blank as histogram specification standard, histogram is carried out to low-quality paper image
Figure specification processing: if the histogram of the histogram distribution of the window of paper image to be processed and former paper image respective window
It is distributed close, then the image to the window is not dealt with, otherwise carries out histogram specification.
6. a kind of mobile phone photograph low-quality images according to claim 1 automatically process and methods of marking, which is characterized in that
It is described that the processing of low-quality images deblurring is specifically comprised the following steps:
Motion blur is eliminated using based on the algorithm of brightness of image and gradient priori, is enabled:
F (x)=| | x | |0
Wherein x indicates that image, f (x) indicate non-zero pixels number in image,
Using priori knowledge, have:
In formulaIndicating image gradient, σ is parameter,
F (x) is added in traditional deblurring model as regularization term, it may be assumed that
Wherein x indicates clear image, and k is fuzzy core, and λ is parameter；
It solves the model and obtains the image after deblurring, method for solving is as follows:
X, k are separated first, obtained:
And
Available x and fuzzy core k is solved for being 1. 2. iterated after above-mentioned separation, wherein every time when iterative solution,
Fixed one of another parameter of parametric solution.
7. a kind of mobile phone photograph low-quality images according to claim 4 automatically process and methods of marking, which is characterized in that
The accurate positionin black square locating piece position in the image denoised with integral nomography specifically comprises the following steps:
The integrogram of rectangular window: defining the black picture element point value in bianry image is 1, and white pixel point value is 0, rectangular window
Integrogram in i (x, y) at any point (x, y) indicate the sum of this upper left corner all pixels in original image:
From top to bottom from given image, the integral image formula being from left to right calculated is as follows:
I (x, y)=i (x-1, y)+i (x, y)+i (x, y-1)-i (x-1, y-1)
After obtaining integrogram, the pixel of the black pixel point in image in any rectangular area and calculated by following formula:
Sum (m, n)=i (x, y)+i (u, v)-i (x, v)-i (u, y)
Wherein rectangle size is m=x-u, and n=y-v, upper left corner of the rectangle coordinate is (u, v), and bottom right angular coordinate is (x, y).
8. a kind of mobile phone photograph low-quality images according to claim 4 automatically process and methods of marking, which is characterized in that
It is described that image progress deformation recovery is included the following steps: according to tag block position
Perspective transform: the perspective transform of image is turned to formula form:
In formula, (x ', y ') indicate that the picture point coordinate after deformation, (x, y) indicate the coordinate of corresponding points in original image.A~l is
Unknown parameter,
Above formula is expressed as matrix multiple:
There are eight unknown parameter a~f in the formula, m, l need to solve, and before being corrected image, need to obtain in original image
Four tag block coordinates and deformation after corresponding four tag block coordinates in image, eight obtained coordinates are denoted as:
Coordinate points in original image: (x1, y1), (x2, y2), (x3, y3), (x4, y4), the coordinate points after deformation: (x1 ', y1 '), (x2 ',
Y2 '), (x3 ', y3 '), (x4 ', y4 '), then available:
The parameter that perspective transform can be obtained is solved, perspective transform correction can be carried out to the image of deformation after obtaining parameter, it can
To obtain unstrained image.
9. a kind of mobile phone photograph low-quality images according to claim 1 automatically process and methods of marking, which is characterized in that
Step 4 specifically comprises the following steps:
Answer region segmentation: the paper image after restoring to deformation is split according to certain rule, obtains the answer area of piecemeal
Domain；
Divide in answer region: for being divided into several rows, the gap-filling questions of ordered series of numbers are answered according to the underscore at its corresponding answer
Topic region is split, and obtains answer content.
10. a kind of mobile phone photograph low-quality images according to claim 9 automatically process and methods of marking, feature exist
In the answer region segmentation specifically comprises the following steps:
Every page of right boundary is estimated according to every page of rectangular locating piece；
Answer region up-and-down boundary is positioned according to the frame of overstriking up and down；
Paper image is cut according to obtained zone boundary.
11. a kind of mobile phone photograph low-quality images according to claim 9 automatically process and methods of marking, feature exist
In dividing in the answer region and specifically comprise the following steps:
Every page of right boundary is estimated according to every page of rectangular locating piece；
Answer region up-and-down boundary is positioned according to the frame of overstriking up and down；
Divided in order according to blank coil information in database and obtained zone boundary information.
12. a kind of mobile phone photograph low-quality images according to claim 1 automatically process and methods of marking, feature exist
In step 5 specifically comprises the following steps:
Position application area position in image；
According to position is smeared, corresponding answer is converted to according to certain rule；
Model answer in comparison database, is scored.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201811596737.9A CN109741273A (en) | 2018-12-26 | 2018-12-26 | A kind of mobile phone photograph low-quality images automatically process and methods of marking |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201811596737.9A CN109741273A (en) | 2018-12-26 | 2018-12-26 | A kind of mobile phone photograph low-quality images automatically process and methods of marking |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN109741273A true CN109741273A (en) | 2019-05-10 |
Family
ID=66361259
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201811596737.9A Pending CN109741273A (en) | 2018-12-26 | 2018-12-26 | A kind of mobile phone photograph low-quality images automatically process and methods of marking |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN109741273A (en) |
Cited By (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110188756A (en) * | 2019-05-31 | 2019-08-30 | 广东利元亨智能装备股份有限公司 | Product positioning method |
| CN110335246A (en) * | 2019-05-29 | 2019-10-15 | 成都数之联科技有限公司 | A kind of license picture clarity evaluation method |
| CN112529822A (en) * | 2021-02-09 | 2021-03-19 | 斯伦贝谢油田技术（山东）有限公司 | Logging-while-drilling imaging data processing method |
| CN113673405A (en) * | 2021-08-14 | 2021-11-19 | 深圳市快易典教育科技有限公司 | Exercise correction method and system based on question recognition and intelligent home teaching and learning machine |
| CN114240765A (en) * | 2021-11-15 | 2022-03-25 | 同济大学 | Method and device for removing interference illumination based on gray-scale camera and storage medium |
Citations (11)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN102360419A (en) * | 2011-09-28 | 2012-02-22 | 广东启明科技发展有限公司 | Method and system for computer scanning reading management |
| CN102760225A (en) * | 2011-04-29 | 2012-10-31 | 贵州师范大学 | Method for positioning answer sheet of test paper objective questions based on rectangular bounding box |
| CN102760226A (en) * | 2011-04-29 | 2012-10-31 | 贵州师范大学 | Solid square-based test paper objective item answer sheet locating method |
| CN104134072A (en) * | 2014-07-04 | 2014-11-05 | 北京学信速达科技有限公司 | Answer sheet identification method |
| CN104915667A (en) * | 2015-05-27 | 2015-09-16 | 华中科技大学 | Mobile-terminal-based method and system for identification and analysis of answering card |
| CN105590101A (en) * | 2015-12-28 | 2016-05-18 | 杭州淳敏软件技术有限公司 | Hand-written answer sheet automatic processing and marking method and system based on mobile phone photographing |
| CN106651689A (en) * | 2016-10-11 | 2017-05-10 | 深圳万发创新进出口贸易有限公司 | Intelligent examination system |
| CN108171297A (en) * | 2018-01-24 | 2018-06-15 | 谢德刚 | A kind of answer card identification method and device |
| US20180247553A1 (en) * | 2017-02-27 | 2018-08-30 | Ricoh Company, Ltd. | Information processing device, non-transitory computer program product, and information processing system |
| WO2018153348A1 (en) * | 2017-02-26 | 2018-08-30 | 朱晓庆 | Recognizable electronic paper and answer sheet |
| CN109033046A (en) * | 2018-06-25 | 2018-12-18 | 陕西师范大学 | Structuring visible document snap information input system and method |
-
2018
- 2018-12-26 CN CN201811596737.9A patent/CN109741273A/en active Pending
Patent Citations (11)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN102760225A (en) * | 2011-04-29 | 2012-10-31 | 贵州师范大学 | Method for positioning answer sheet of test paper objective questions based on rectangular bounding box |
| CN102760226A (en) * | 2011-04-29 | 2012-10-31 | 贵州师范大学 | Solid square-based test paper objective item answer sheet locating method |
| CN102360419A (en) * | 2011-09-28 | 2012-02-22 | 广东启明科技发展有限公司 | Method and system for computer scanning reading management |
| CN104134072A (en) * | 2014-07-04 | 2014-11-05 | 北京学信速达科技有限公司 | Answer sheet identification method |
| CN104915667A (en) * | 2015-05-27 | 2015-09-16 | 华中科技大学 | Mobile-terminal-based method and system for identification and analysis of answering card |
| CN105590101A (en) * | 2015-12-28 | 2016-05-18 | 杭州淳敏软件技术有限公司 | Hand-written answer sheet automatic processing and marking method and system based on mobile phone photographing |
| CN106651689A (en) * | 2016-10-11 | 2017-05-10 | 深圳万发创新进出口贸易有限公司 | Intelligent examination system |
| WO2018153348A1 (en) * | 2017-02-26 | 2018-08-30 | 朱晓庆 | Recognizable electronic paper and answer sheet |
| US20180247553A1 (en) * | 2017-02-27 | 2018-08-30 | Ricoh Company, Ltd. | Information processing device, non-transitory computer program product, and information processing system |
| CN108171297A (en) * | 2018-01-24 | 2018-06-15 | 谢德刚 | A kind of answer card identification method and device |
| CN109033046A (en) * | 2018-06-25 | 2018-12-18 | 陕西师范大学 | Structuring visible document snap information input system and method |
Non-Patent Citations (1)
| Title |
|---|
| 张红: ""基于L_0正则化的文本图像去模糊方法研究"", 《中国优秀硕士学位论文全文数据库 信息科技辑》 * |
Cited By (11)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110335246A (en) * | 2019-05-29 | 2019-10-15 | 成都数之联科技有限公司 | A kind of license picture clarity evaluation method |
| CN110335246B (en) * | 2019-05-29 | 2021-04-13 | 成都数之联科技有限公司 | License picture definition evaluation method |
| CN110188756A (en) * | 2019-05-31 | 2019-08-30 | 广东利元亨智能装备股份有限公司 | Product positioning method |
| WO2020238116A1 (en) * | 2019-05-31 | 2020-12-03 | 广东利元亨智能装备股份有限公司 | Product positioning method |
| US20220076428A1 (en) * | 2019-05-31 | 2022-03-10 | Guangdong Lyric Robot Automation Co., Ltd. | Product positioning method |
| US12131512B2 (en) * | 2019-05-31 | 2024-10-29 | Guangdong Lyric Robot Automation Co., Ltd. | Product positioning method |
| CN112529822A (en) * | 2021-02-09 | 2021-03-19 | 斯伦贝谢油田技术（山东）有限公司 | Logging-while-drilling imaging data processing method |
| CN112529822B (en) * | 2021-02-09 | 2021-04-20 | 斯伦贝谢油田技术（山东）有限公司 | Logging-while-drilling imaging data processing method |
| CN113673405A (en) * | 2021-08-14 | 2021-11-19 | 深圳市快易典教育科技有限公司 | Exercise correction method and system based on question recognition and intelligent home teaching and learning machine |
| CN113673405B (en) * | 2021-08-14 | 2024-03-29 | 深圳市快易典教育科技有限公司 | Exercise correction method, system and intelligent tutor learning machine based on topic recognition |
| CN114240765A (en) * | 2021-11-15 | 2022-03-25 | 同济大学 | Method and device for removing interference illumination based on gray-scale camera and storage medium |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN109741273A (en) | A kind of mobile phone photograph low-quality images automatically process and methods of marking | |
| EP3309703B1 (en) | Method and system for decoding qr code based on weighted average grey method | |
| CN112183038A (en) | Form identification and typing method, computer equipment and computer readable storage medium | |
| CN107247950A (en) | A kind of ID Card Image text recognition method based on machine learning | |
| CN110097046A (en) | A kind of character detecting method and device, equipment and computer readable storage medium | |
| US20120099792A1 (en) | Adaptive optical character recognition on a document with distorted characters | |
| US11836969B2 (en) | Preprocessing images for OCR using character pixel height estimation and cycle generative adversarial networks for better character recognition | |
| CN104463795A (en) | Processing method and device for dot matrix type data matrix (DM) two-dimension code images | |
| CN108108734B (en) | License plate recognition method and device | |
| CN106875546A (en) | A method for identifying value-added tax invoices | |
| CN110807454B (en) | Text positioning method, device, equipment and storage medium based on image segmentation | |
| CN112507782A (en) | Text image recognition method and device | |
| CN110598566A (en) | Image processing method, device, terminal and computer readable storage medium | |
| CN111507181B (en) | Correction method and device for bill image and computer equipment | |
| CN111046644A (en) | Answer sheet template generation method, identification method, device and storage medium | |
| CN113139535A (en) | OCR document recognition method | |
| Lu et al. | A shadow removal method for tesseract text recognition | |
| EP2545498B1 (en) | Resolution adjustment of an image that includes text undergoing an ocr process | |
| CN105551044A (en) | Picture comparing method and device | |
| CN111274863A (en) | Text prediction method based on text peak probability density | |
| CN114267035A (en) | Document image processing method and system, electronic device and readable medium | |
| Bala et al. | Image simulation for automatic license plate recognition | |
| WO2022056875A1 (en) | Method and apparatus for segmenting nameplate image, and computer-readable storage medium | |
| CN109635798B (en) | Information extraction method and device | |
| US9269126B2 (en) | System and method for enhancing the legibility of images |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| TA01 | Transfer of patent application right |
Effective date of registration: 20201130 Address after: No. 159, Chengjiang Middle Road, high tech Zone, Jiangyin City, Wuxi City, Jiangsu Province Applicant after: Jiangsu youlixin Education Technology Co., Ltd Address before: 214400 A604, 159 Chengjiang Middle Road, Jiangyin High-tech Zone, Wuxi City, Jiangsu Province Applicant before: JIANGSU YOUSHENG INFORMATION TECHNOLOGY Co.,Ltd. |
|
| TA01 | Transfer of patent application right | ||
| WD01 | Invention patent application deemed withdrawn after publication |
Application publication date: 20190510 |
|
| WD01 | Invention patent application deemed withdrawn after publication |