CN111680549B - A paper pattern identification method - Google Patents

Abstract

The invention discloses a paper grain identification method, which includes the following steps: S1. Under transmitted light source conditions, take microscopic images with micron precision of the internal fibers of the reference paper document and the paper document to be identified, and use them as reference paper grain images respectively. and the paper texture image to be identified; S2. Extract and match the feature points of the reference paper texture image and the paper texture image to be identified, and generate a feature point matching pair; S3. Estimate the reference paper texture image based on the feature point matching pair. and the transformation matrix between the paper texture image to be identified, and obtain the areas of interest of the reference paper texture image and the paper texture image to be identified respectively; S4. Enhance the fiber texture of the area of interest respectively; S5. The similarity is measured based on the enhanced texture structure and the recognition result is output. This method can not only resist the translation, rotation, and scaling of the paper pattern image caused by illumination changes or human and equipment operation deviations during the paper pattern collection process, but can also resist abnormal situations such as stolen goods appearing on the surface of the paper.

Description

A paper pattern identification method

Technical field

The invention relates to the field of object identification, and in particular to a paper pattern identification method based on the microscopic texture structure of the paper itself.

Background technique

With the rapid development of computer hardware and computer vision technology, more image processing-based technologies have received widespread attention in the field of anti-counterfeiting identification of items. Regarding the anti-counterfeiting issues of important paper documents, such as contracts, bills, performance tickets, etc., due to the lack of significant marking features of such documents, and the development of today's printing technology and printing accuracy, the difficulty of counterfeiting them has been greatly reduced. Not only The technical threshold for counterfeiting is low, and the counterfeit copies are also very low, which makes them a key target of some counterfeiters.

Most of the traditional methods are to mark and prevent counterfeiting of important documents by signing, stamping, printing anti-counterfeiting labels, etc. Although these methods are low-cost and easy to implement, they are prone to counterfeit attacks and do not have good anti-counterfeiting performance. In addition, there are anti-counterfeiting technologies such as the use of special anti-counterfeiting paper, special inks, and the artificial addition of random fibers and random bubble anti-counterfeiting labels to items through physical or chemical methods. However, these methods have high costs and technical barriers, which are not conducive to promotion and application. Generally, only For packaging of expensive goods to prevent counterfeiting. In recent years, due to the rapid development of image acquisition, image processing and computer technology, an identification method based on the natural unclonable characteristics of paper itself has become a hot issue;

In particular, patent CN 102073865 A proposes an anti-counterfeiting method that utilizes the fiber texture of paper itself, which can achieve anti-counterfeiting recognition of paper without additional technical processing of the paper. However, this method is based on pattern recognition of extracted texture feature points. The results determine the authenticity of the paper, and the extracted feature points depend on the position of the paper during each collection, the relative position of the collection equipment and the paper, and the optical magnification of the collection equipment; therefore, in order to ensure the accuracy of the recognition results, it is required to collect images when The above collection conditions are highly consistent and cannot resist the translation, rotation and scaling of the paper. In addition, when contaminants appear on the surface of the paper, such as writing, water stains, etc., the extracted feature points will also change, resulting in paper recognition results. Inaccurate;

Patent CN 102955930 A describes an anti-counterfeiting method that uses the physical characteristics of a substance to identify itself. Compared with the above-mentioned patent CN 102073865 A, the difference is that the substance is light-transmissive, and pattern recognition is performed by obtaining the physical characteristics image collected optically after transmission. ; The purpose of using transmitted light is to realize image acquisition of its physical characteristics at low cost, but it still has the problems of patent CN 102073865 A;

A paper texture identification method described in patent CN 110599665 A. Compared with patent CN 102073865 A, before extracting the feature points of the paper texture image, the decontamination feature area image of the paper texture to be verified is obtained through the Yolo-v2 model. Due to this The final identification of the identification method is still based on the feature vector of the feature points of the collected paper texture image, the identification result is the location of the contaminated area, the area size, the location distribution of the remaining area after the contamination is removed, and the distribution of feature points in the remaining area after the contamination is removed. Impact;

In summary, the existing paper pattern recognition methods based on the natural unclonable characteristics of paper patterns have two obvious limitations. First, the paper texture recognition method cannot resist the translation, rotation and scaling of the paper texture image caused by human and equipment operation deviations during paper texture collection. Second, the paper texture recognition method cannot resist the dirt on the paper surface, and the recognition results will be stolen. Great impact.

Contents of the invention

The technical problem to be solved by the present invention is that in view of the shortcomings of the existing technology, the present invention provides a new paper pattern recognition method by combining the paper pattern collection scheme and the paper pattern recognition algorithm.

The technical solution adopted by the present invention to solve the technical problem is to construct a paper pattern recognition method, which includes the following steps:

S1. Under transmitted light source conditions, take micron-precision microscopic images of the internal fibers of the reference paper document and the paper document to be identified, respectively, as the reference paper grain image and the paper grain image to be identified;

S2. Extract and match the feature points of the reference paper texture image and the paper texture image to be identified, and generate a feature point matching pair;

S3. Estimate the transformation matrix between the reference paper texture image and the paper texture image to be identified according to the feature point matching pair, and obtain the areas of interest of the reference paper texture image and the paper texture image to be identified respectively;

S4. Enhance the fiber texture of the area of interest respectively;

S5. Measure the similarity based on the enhanced texture structure and output the recognition result.

Preferably, in the paper texture identification method of the present invention, in step S1,

When collecting the reference paper grain image, simultaneously record the coordinate position of the corresponding point of the center point of the collected reference paper grain image on the reference paper document relative to the center point of the reference paper document;

When collecting the paper grain image to be identified, the paper grain image to be identified in the corresponding area is collected according to the recorded coordinate position.

Preferably, in the paper grain identification method of the present invention, when the paper grain image to be identified is collected, the recorded coordinate position is within the field of view of the microscopic photography.

Preferably, in the paper grain identification method of the present invention, when the focus is accurate, microscopic images of the reference paper document and the paper document to be identified are taken.

Preferably, in the paper texture identification method of the present invention, the step S2 includes:

S2-1. Perform image preprocessing on the reference paper texture image and the paper texture image to be identified;

S2-2. Extract texture feature points from the preprocessed reference paper texture image and the paper texture image to be identified, calculate feature vectors of feature points, match feature points according to the feature vectors, and generate feature point matching pairs.

Preferably, in the paper pattern recognition method of the present invention, matching feature points according to feature vectors includes:

Feature points are matched according to the similarity of the feature vectors corresponding to the extracted feature points.

Preferably, in the paper pattern recognition method of the present invention, matching feature points according to the similarity of feature vectors corresponding to the extracted feature points includes:

Extract the feature points of the reference paper texture image and the paper texture image to be identified respectively, calculate the similarity of the feature vectors corresponding to the feature points of the reference paper texture image and the paper texture image to be identified, respectively, and select a threshold value. When the calculated similarity is greater than the threshold, a matching pair of feature points is generated.

Preferably, in the paper texture identification method of the present invention, the feature points are SURF or SIFT feature points.

Preferably, in the paper texture identification method of the present invention, the step S3 includes:

S3-1. Eliminate erroneous feature point matching pairs among the feature point matching pairs obtained in step S2, and estimate the relationship between the reference paper texture image and the paper texture image to be identified based on the remaining valid feature point matching pairs. transformation matrix;

S3-2. Use the transformation matrix to transform the paper texture image to be identified;

S3-3. Overlay the transformed paper texture image to be identified and the reference paper texture image, and intercept the inscribed rectangles of the overlapping parts from the two paper texture images as regions of interest.

Preferably, in the paper texture identification method of the present invention,

The step S3-1 includes: using the MSAC algorithm to eliminate erroneous feature point matching pairs among the feature point matching pairs obtained in the step S2, using the reference paper texture image as a standard, and based on the remaining valid feature point matching pairs, Estimate the affine transformation matrix from the paper texture image to be identified to the reference paper texture image;

The step S3-2 includes: using the affine transformation matrix to perform affine transformation on the paper texture image to be identified;

The step S3-3 includes: stacking the transformed paper texture images, selecting a threshold S, and comparing the area of the overlapping portion after the transformation to the total area of the reference paper texture image or the paper texture image to be identified. When it is greater than S, the largest inscribed rectangle of the overlapping part after transformation is intercepted as the area of interest.

Preferably, in the paper texture identification method of the present invention, step S3 further includes:

If the number of effective feature point matching pairs in step S3-1 is too small to estimate the transformation matrix, or the area of interest area is too small, then the reference paper texture preprocessed in step S2-1 is used. image and the paper texture image to be identified as the area of interest.

Preferably, in the paper texture identification method of the present invention, the step S4 includes:

Gabor filters at multiple angles are used to enhance the fiber texture of the two regions of interest respectively. At the same time, the Gabor filter at each angle generates an amplitude response matrix and a phase angle response matrix, and then the Gabor filters at multiple angles are The amplitude response matrix and the phase angle response matrix are superimposed, and finally a corresponding amplitude response matrix and phase angle response matrix are generated.

Preferably, in the paper texture identification method of the present invention, the step S4 includes:

According to the image size parameters given in advance, the areas of interest of the reference paper texture image and the paper texture image to be identified are adjusted to the same size.

Preferably, in the paper texture identification method of the present invention, the step S5 includes:

Map the enhanced texture structure of the area of interest of the paper texture image to be identified and the reference paper texture image to 01 digital space respectively, and use Hamming distance to generate a similarity index to measure the reference paper texture image and the paper texture to be identified. Image correspondence to the similarity of the generated bitstream.

Preferably, in the paper pattern recognition method of the present invention, the enhanced texture structure is mapped to 01 digital space, and the Hamming distance is used to generate a similarity index to measure the reference paper pattern image and the to-be-identified The similarities of the bitstream generated corresponding to the paper texture image include:

Calculate the mean of the final amplitude response matrix and phase angle response matrix of the reference paper texture image and the area of interest of the paper texture image to be identified respectively, compare each number in the response matrix with the mean, and compare the mean The larger one is 1, and the smaller one is 0. After expanding the two digital matrices by row or column, the resulting bit streams are spliced together and used as the reference paper texture image and the paper texture image to be identified respectively. digital paper texture;

Calculate the Hamming distance between the reference paper texture image and the digital paper texture of the paper texture image to be identified, and use the proportion of the Hamming distance to the total length of the digital paper texture as the distance between the reference paper texture and the paper texture to be identified. similarity index;

Select a threshold t. If the similarity index is greater than the threshold t, the recognition fails. If the similarity index is less than the threshold t, the recognition is successful.

The present invention proposes a paper pattern recognition method based on the micron-precision microstructure of fibers inside the paper. It mainly uses the texture structure of the random microfiber interweaving inside the paper to identify the paper pattern. Compared with the existing technology, the paper pattern identification method proposed by the present invention has at least the following beneficial effects: (1) Using a paper pattern collection scheme using microscopy and transmitted illumination, it can directly collect the paper surface to the following 50 micron range Micron-precision fiber structure texture image, and the fiber texture is stable and insensitive to changes in illumination and paper surface dirt; (2) Using micron-precision fiber texture characteristics of paper for paper pattern recognition ensures that the feature point prediction method proposed by the present invention is Estimating the stability of the paper texture calibration method of the paper texture transformation matrix can resist the translation, rotation and scaling of the paper texture image caused by illumination changes or human and equipment operation deviations during the paper texture collection process; (3) Due to micron precision fibers Texture features are not easily lost when the paper surface is dirty, ensuring that the paper pattern recognition algorithm proposed in the present invention can still achieve paper texture calibration and recognition work in the case of stolen goods, so that the paper pattern recognition method proposed in the present invention can combat stolen goods on the paper surface. and other unfavorable situations.

Description of drawings

The present invention will be further described below in conjunction with the accompanying drawings and examples. In the accompanying drawings:

Figure 1 is a flow chart of the paper pattern identification method of the present invention;

Figure 2 is a flow chart of the present invention for obtaining the reference paper texture image and the area of interest of the paper texture image to be identified;

Figure 3 is a schematic diagram of an example of the area of interest of the reference paper texture image and the paper texture image to be identified according to the present invention;

Figure 4 is a schematic diagram of an example of using Gabor filters at multiple angles to enhance the fiber texture in the area of interest.

Detailed ways

In order to have a clearer understanding of the technical features, purposes and effects of the present invention, the specific embodiments of the present invention will now be described in detail with reference to the accompanying drawings.

As shown in Figure 1, the present invention constructs a paper texture recognition method, which is based on the microscopic texture structure of the internal fibers of the paper. It is easy to implement, does not require the introduction of additional printing operations, and can resist paper translation, rotation, and image scaling during collection. , and at the same time, it is possible to ensure the accuracy of the results of the paper pattern identification method when different contaminants appear on the surface of the paper. The method includes the following steps:

Step S1: Under transmitted light source conditions, take microscopic images with micron precision of the internal fibers of the reference paper document and the paper document to be identified, respectively, as the reference paper grain image and the paper grain image to be identified.

Specifically, in step S1, the reference paper texture image needs to be collected and stored in advance for subsequent paper texture recognition. The paper texture image to be identified is used to identify and match the reference paper texture image. When the paper texture image to be identified and the reference paper texture image are collected from the same area of the same paper, the paper texture matching is successful; when the paper texture image to be identified and the reference paper texture image are When paper texture images are collected from different areas of the same paper or different papers, paper texture matching fails.

The above-mentioned same area of the same paper means that the collected reference paper texture image and the paper texture image to be identified contain a common area on the same paper document, but it does not require that the corresponding areas of the two collected images are exactly the same, that is, the collected paper Translation and rotation caused by manual acquisition are allowed when texture images are taken.

Among them, step S1 specifically includes: when collecting the reference paper grain image, simultaneously recording the coordinate position of the corresponding point of the center point of the collected reference paper grain image on the reference paper document relative to the center point of the reference paper document; when collecting the reference paper grain image to be identified When extracting a paper texture image, collect the paper texture image to be identified in the corresponding area according to the recorded coordinate position.

Moreover, when collecting the paper pattern image to be identified, it is only necessary to ensure that the recorded coordinate position relative to the center point of the reference paper document is within the field of view of the microscope photography.

In addition, when collecting microscopic images of the reference paper document and the paper document to be identified, you only need to ensure that the focus is accurate during each collection. It does not require that the magnification when collecting the image is highly consistent, which means that the reference paper texture can be collected. The magnification of the image and the paper texture image to be recognized are inconsistent. Therefore, in this embodiment, when the focus is accurate, microscopic images of the reference paper document and the paper document to be identified are captured.

The acquisition scheme of microscopy and transmitted illumination described in step S1 can collect fiber images with micron precision of the fibers inside the paper. The texture features are stable and insensitive to abnormal conditions such as changes in illumination and dirt on the paper surface.

Step S2: Extract and match the feature points of the reference paper texture image and the paper texture image to be identified, and generate a feature point matching pair.

Specifically, step S2 includes:

Step S2-1: Perform image preprocessing on the reference paper texture image and the paper texture image to be identified;

Among them, the purpose of image preprocessing is to simplify the reference paper texture image and the paper texture image data to be identified, and make them have the same size and data format, which can effectively avoid recognition errors caused by inconsistent image data formats. If the collected image With multiple channels, select one of the channel data for processing. Specifically, the reference paper grain image and the paper grain image to be identified are grayscaled and adjusted to the same size. In this embodiment, the reference paper grain image and the paper grain image to be identified are grayscaled and adjusted to 640 *640 size.

Step S2-2: Extract texture feature points from the preprocessed reference paper texture image and the paper texture image to be identified, calculate the feature vectors of the feature points, match the feature points according to the feature vectors, and generate matching pairs of feature points.

Among them, feature points generally refer to points that are invariant to rotation, translation, or affine transformation. The feature matching method can effectively resist the translation, rotation, and miscut of the collected images, and also resist illumination changes, noise, and viewpoint transformations. Insensitive and improve algorithm robustness.

In this embodiment, matching the feature points according to the feature vectors includes: matching the feature points according to the similarity of the feature vectors corresponding to the extracted feature points. Specifically, the feature points of the reference paper texture image and the paper texture image to be identified are respectively extracted, the similarity of the feature vectors corresponding to the feature points of the reference paper texture image and the paper texture image to be identified is calculated respectively, and a threshold is selected. When the calculated similarity When it is greater than the threshold, a matching pair of feature points is generated. In some embodiments, the feature points are SURF or SIFT feature points. In this embodiment, the feature points are SURF feature points, which can effectively resist translation, rotation, scaling, and illumination changes when collecting paper texture images.

Compared with the existing technology, the matching feature point step described in step S2 is to prepare for estimating the transformation matrix in step S3 and obtaining the reference paper texture image and the area of interest of the paper texture image to be identified, rather than using Paper texture recognition is directly performed based on the feature vectors of the feature points.

Step S3: Estimate the transformation matrix between the reference paper texture image and the paper texture image to be identified based on the feature point matching pair, and obtain the areas of interest of the reference paper texture image and the paper texture image to be identified respectively;

Specifically, step S3 includes:

Step S3-1: Eliminate erroneous feature point matching pairs among the feature point matching pairs obtained in step S2, and estimate the transformation matrix between the reference paper texture image and the paper texture image to be identified based on the remaining valid feature point matching pairs, such that After the transformation, the corresponding effective feature point matching pairs between the two are as consistent as possible in spatial position;

Step S3-2: Use the transformation matrix to transform the paper texture image to be identified;

Step S3-3: Overlay the transformed paper texture image to be identified and the reference paper texture image, and intercept the inscribed rectangles of the overlapping parts from the two paper texture images as regions of interest to achieve paper texture calibration. There are two areas of interest here, one is the reference paper texture image, and the other is the paper texture image to be recognized.

The purpose of eliminating matching pairs of feature points described in step S3-1 is to effectively reduce the impact of noise on feature point selection, eliminate erroneous matching pairs of feature points caused by noise interference, and ensure that the estimated value to be estimated in step S3-1 is Accuracy in identifying transformation matrices between paper texture images and reference paper texture images.

The transformation matrix described in step S3-1, when the reference paper texture image and the paper texture image to be identified are collected from the same area of the same paper, can automatically adjust the position translation of the paper texture to be identified relative to the reference paper texture, Rotate and scale.

The area of interest described in step S3-1, when the reference paper texture image and the paper texture image to be identified are collected from the same area of the same piece of paper, the obtained area of interest is the common area of the two collections. .

In this embodiment, preferably, step S3-1 includes: using the MSAC algorithm to eliminate erroneous feature point matching pairs from the feature point matching pairs obtained in step S2, using the reference paper texture image as the standard, and based on the remaining valid feature points Matching pairs, estimating the affine transformation matrix from the paper texture image to be recognized to the reference paper texture image;

Step S3-2 includes: using an affine transformation matrix to perform affine transformation on the paper texture image to be identified, so that the corresponding effective feature point matching pairs between the two are as consistent as possible in spatial position after transformation;

Step S3-3 includes: overlaying the transformed paper texture image to be identified and the reference paper texture image, so that the reference paper texture image and the paper texture image to be identified have a larger similar area, and selecting the threshold S. If after transformation When the area of the overlapping part accounts for the total area of the reference paper texture image or the paper texture image to be identified is greater than S, then the largest inscribed rectangle of the overlapping part of the two after transformation is intercepted as the area of interest. In this example, S=20%.

In some embodiments, step S3 also includes: if the number of effective feature point matching pairs in step S3-1 is too small to estimate the transformation matrix, or the area of the region of interest is too small, then use the predetermined method in step S2-1. The processed reference paper texture image and the paper texture image to be identified are used as regions of interest.

Specifically, in this embodiment, since estimating an affine transformation matrix requires three feature point pairs, if the number of remaining valid feature point matching pairs in step S3-1 is less than 3, it is not enough to estimate the affine transformation matrix or After the transformation, the overlapping area ratio of the reference paper texture and the paper texture to be identified is less than 20%, then the reference paper texture image and the paper texture image to be identified that were preprocessed in step S2-1 are used as the area of interest.

Compared with the existing technology, the present invention can achieve registration between the collected paper texture images by intercepting the area of interest without additional operations such as printing rectangles and watermarks.

Compared with the existing technology, the present invention ensures that paper texture recognition does not rely on the extraction of global feature points by intercepting areas of interest. When local graffiti, water stains, or printed text appear on the surface of the paper, no prior decontamination is required. The accuracy of paper grain identification can be guaranteed.

The step S3 uses the feature point matching pairs extracted in the step S2 to obtain the area of interest, which can effectively resist the translation and rotation of the paper, the changes in the lighting of the environment, and the relative position changes between the paper and the image acquisition device when collecting paper textures. As well as changes in the optical magnification of image acquisition, compared with the existing method of directly identifying paper grains by relying on the feature vectors of global feature points, the method proposed by the present invention can well combat the differences caused by inconsistent manual operations when collecting images. .

The process of obtaining the area of interest of the reference paper texture image and the paper texture image to be identified in step S3 is shown in Figure 2. The schematic diagram of an example of obtaining the area of interest of the reference paper texture image and the paper texture image to be identified (reference paper texture and the area of interest to be identified) The paper texture is collected from the same area of the same paper) (see Figure 3;

The stability of the paper texture calibration method based on the inter-image transformation of the paper fiber microscopic image feature point estimation in steps S2 and S3 is guaranteed by the stability of the micron-precision microscopic image texture features of the internal fibers of the paper collected in step S1. The calibration algorithm can resist the translation, rotation and scaling of the paper texture image caused by illumination changes or human and equipment operation deviations during the paper texture collection process.

Step S4: Enhance the fiber texture of the area of interest respectively.

Specifically, step S4 includes:

Gabor filters at multiple angles are used to enhance the fiber texture of the two areas of interest. At the same time, the Gabor filter at each angle generates an amplitude response matrix and a phase angle response matrix, and then the amplitudes at multiple angles are The response matrix and the phase angle response matrix are superimposed, and the last paper texture image generates a corresponding amplitude response matrix and phase angle response matrix.

Existing paper texture anti-counterfeiting technologies mostly use unidirectional Gabor filtering to enhance the paper texture, and then use the eigenvectors or singular values of the generated matrix for final paper texture recognition. This will cause the details of the paper texture to be lost and affect the recognition. accuracy.

The multi-angle Gabor filter enhancement described above, taking into account the randomness of the paper surface and internal microfiber texture arrangement and direction, can retain the paper texture information to a greater extent and improve the recognition accuracy.

Specifically, a multi-angle Gabor filter is used to enhance the fiber texture of the reference paper texture image and the area of interest of the paper texture image to be identified. The Gabor filter has the characteristics of achieving optimal localization in the spatial domain and frequency domain at the same time, so it can easily Good description of local structural information corresponding to spatial frequency, spatial position and direction selectivity. In this embodiment, Gabor filters in four directions are applied to a paper pattern at the same time. The spatial frequencies of the four Gabor filters are set to 0.1, and then the direction parameters are set to 0°, 30°, and 60 respectively. °, 90°, so each paper pattern can obtain four amplitude responses and four phase angle responses corresponding to four Gabor filters. The four amplitude responses and four phase angle responses are superimposed respectively to obtain a joint amplitude response. The value response and phase angle response are the final response results of a piece of paper texture. It should be pointed out that the size of the amplitude response and phase angle response matrix here is the same as the size of the input area of interest. Therefore, in order to ensure that the recognition results of different paper textures are comparable, it is necessary to follow the pre-given image before enhancing the paper texture. The size parameter adjusts the area of interest of the reference paper texture image and the paper texture image to be identified to the same size. Because the overlapping areas of different paper texture images have different sizes, the size of the intercepted area of interest may be different. In this embodiment, before enhancing the texture, the regions of interest are all adjusted to 32*32 size images. Among them, a schematic diagram of an example of using Gabor filters at multiple angles to enhance the fiber texture in the area of interest is shown in Figure 4.

Since the micron-precision microtexture characteristics of the internal fibers of the paper collected by this method are stable and can remain stable under the conditions of changes in illumination and dirt on the surface of the paper, the recognition algorithm based on the paper fiber texture image described in step S4 can be Fight against abnormal conditions such as lighting changes and paper soiling.

Step S5: Measure the similarity based on the enhanced texture structure and output the recognition result.

Specifically, the method of measuring similarity based on the enhanced texture structure described in step S5 is:

The enhanced texture structure of the area of interest of the paper texture image to be identified and the reference paper texture image is mapped to the 01 digital space respectively, and the Hamming distance is used to generate a similarity index to measure the corresponding generated results of the reference paper texture image and the paper texture image to be identified. Bitstream similarity.

Preferably, the enhanced texture structure is mapped to 01 digital space, the Hamming distance is used to generate a similarity index, and the similarity of the bit stream generated corresponding to the reference paper texture image and the paper texture image to be identified includes:

Calculate the mean of the final amplitude response matrix and phase angle response matrix of the reference paper texture image and the area of interest of the paper texture image to be identified respectively, compare each number in the response matrix with the mean, and take 1 if it is larger than the mean. , which is smaller than the mean value, is set to 0. After expanding the two digital matrices by row or column, the resulting bit streams are spliced together and used as the reference paper texture image and the digital paper texture of the paper texture image to be recognized respectively;

Calculate the Hamming distance between the reference paper texture image and the digital paper texture of the paper texture image to be identified, and use the proportion of the Hamming distance to the total length of the digital paper texture as the similarity index between the reference paper texture and the paper texture to be identified;

Select the threshold t. If the similarity index is greater than the threshold t, the recognition fails. If the similarity index is less than the threshold t, the recognition is successful.

Specifically, calculate the mean of the final amplitude response and phase angle response matrix of the area of interest, and then compare each number in the response matrix with the mean. If it is larger than the mean, it will be 1, and if it is smaller than the mean, it will be 0; In this embodiment, for a 32*32 area of interest of a paper pattern, two 32*32 01 digital matrices corresponding to the amplitude response and phase angle response can finally be obtained. The two digital matrices are divided into rows or After expanding by columns, the obtained bit streams are spliced together to obtain a bit stream with a size of 32*32*2=2048, which is used as the reference paper texture image and the digital paper texture of the paper texture image to be recognized respectively.

In information encoding, the number of different bits encoded on the corresponding bits of two legal codes is called the Hamming distance; in this embodiment, the reference paper texture obtained in the above steps is calculated and the digital paper texture of the paper texture image to be recognized Hamming distance between them, and then use the proportion of the Hamming distance to the total length of the digital paper texture 2048 as the similarity index between the reference paper texture and the paper texture to be identified. For example, when the reference paper texture image and the paper texture image to be identified When the number of different digits in the corresponding bits of the generated digital paper texture is 80, that is, when the Hamming distance is 80, the corresponding similarity index is 0.039; it should be pointed out that if the reference paper texture image and the paper texture image to be identified Collected from the same area of the same paper, the similarity index between the reference paper texture image and the digital paper texture corresponding to the paper texture image to be identified is close to 0; if the reference paper texture image and the paper texture image to be identified are collected from different papers, or they are the same In different areas of the paper, due to randomness, the similarity index between the reference paper texture image and the digital paper texture corresponding to the paper texture image to be identified is close to 0.5. In this embodiment, the threshold t=0.25 is selected. When the similarity index is greater than 0.25, the recognition fails. When the similarity index is less than 0.25, the recognition is successful. Moreover, it should be noted that the selection of the threshold can be finally determined based on a large number of experimental results.

The present invention proposes a paper pattern identification method based on the micron-precision fiber microstructure inside the paper, which mainly uses the random microfiber interweaving texture structure inside the paper to identify the paper pattern. Compared with the existing technology, the paper pattern identification method proposed by the present invention has at least the following beneficial effects: (1) Using a paper pattern collection scheme using microscopy and transmitted illumination, it can directly collect the paper surface to the following 50 micron range Micron-precision fiber structure texture image, and the fiber texture is stable and insensitive to changes in illumination and paper surface dirt; (2) Using micron-precision fiber texture characteristics of paper for paper pattern recognition ensures that the feature point prediction method proposed by the present invention is Estimating the stability of the paper texture calibration method of the paper texture transformation matrix can resist the translation, rotation and scaling of the paper texture image caused by human and equipment operation deviations in the paper texture collection process; (3) Due to the fiber texture characteristics of micron precision in It is not easy to lose when the surface of the paper is dirty, ensuring that the paper pattern recognition algorithm proposed in the present invention can still achieve paper texture calibration and recognition work in the case of stolen goods, so that the paper pattern recognition method proposed in the present invention can resist unfavorable situations such as stolen goods on the paper surface. .

The present invention is illustrated through specific embodiments. Those skilled in the art will understand that various transformations and equivalent substitutions can be made to the present invention without departing from the scope of the present invention. In addition, various modifications may be made to adapt the invention to a particular situation or situation without departing from the scope of the invention. Therefore, the present invention is not limited to the specific embodiments disclosed, but it is intended to include all embodiments falling within the scope of the claims of the present invention.

Claims (12)

1. A paper pattern identification method, characterized in that it includes the following steps: S1. Under transmitted light source conditions, take microscopic images with micron precision of the internal fibers of the reference paper document and the paper document to be identified, respectively, as the reference paper grain image and the paper grain image to be identified; S2. Extract and match the feature points of the reference paper texture image and the paper texture image to be identified, and generate a feature point matching pair; S3. Estimate the transformation matrix between the reference paper texture image and the paper texture image to be identified according to the feature point matching pair, and obtain the areas of interest of the reference paper texture image and the paper texture image to be identified respectively; S4. Enhance the fiber texture of the area of interest respectively; S5. Measure the similarity based on the enhanced texture structure and output the recognition result; Among them, the step S2 includes: S2-1. Perform image preprocessing on the reference paper texture image and the paper texture image to be identified; S2-2. Extract and match the feature points of the preprocessed reference paper texture image and the paper texture image to be identified, and generate a feature point matching pair; The step S3 includes: S3-1. Eliminate erroneous feature point matching pairs among the feature point matching pairs obtained in step S2, and estimate the relationship between the reference paper texture image and the paper texture image to be identified based on the remaining valid feature point matching pairs. transformation matrix; S3-2. Use the transformation matrix to transform the paper texture image to be identified; S3-3. Overlay the transformed paper texture image to be identified and the reference paper texture image, and intercept the inscribed rectangles of the overlapping parts from the two paper texture images as regions of interest; The step S3 also includes: If the number of effective feature point matching pairs in step S3-1 is too small to estimate the transformation matrix, or the area of interest area is too small, then the reference paper texture preprocessed in step S2-1 is used. The image and the paper texture image to be identified are used as the area of interest; The step S4 includes: Gabor filters at multiple angles are used to enhance the fiber texture of the two regions of interest respectively. At the same time, the Gabor filter at each angle generates an amplitude response matrix and a phase angle response matrix, and then the Gabor filters at multiple angles are The amplitude response matrix and the phase angle response matrix are superimposed, and finally a corresponding amplitude response matrix and phase angle response matrix are generated. 2. The paper texture identification method according to claim 1, characterized in that, in step S1, When collecting the reference paper grain image, simultaneously record the coordinate position of the corresponding point of the center point of the collected reference paper grain image on the reference paper document relative to the center point of the reference paper document; When collecting the paper grain image to be identified, the paper grain image to be identified in the corresponding area is collected according to the recorded coordinate position. 3. The paper pattern identification method according to claim 2, characterized in that when the paper pattern image to be identified is collected, the recorded coordinate position is within the field of view of the microscopic photography. 4. The paper pattern identification method according to claim 1, characterized in that, when the focus is accurate, microscopic images of the reference paper document and the paper document to be identified are taken. 5. The paper pattern identification method according to claim 1, characterized in that the step S2-2 includes: Texture feature points are extracted from the preprocessed reference paper texture image and the paper texture image to be identified, feature vectors of the feature points are calculated, feature points are matched according to the feature vectors, and feature point matching pairs are generated. 6. The paper pattern recognition method according to claim 5, characterized in that matching feature points according to feature vectors includes: Feature points are matched according to the similarity of the feature vectors corresponding to the extracted feature points. 7. The paper pattern recognition method according to claim 6, wherein matching feature points according to the similarity of feature vectors corresponding to the extracted feature points includes: Extract the feature points of the reference paper texture image and the paper texture image to be identified respectively, calculate the similarity of the feature vectors corresponding to the feature points of the reference paper texture image and the paper texture image to be identified, respectively, and select a threshold value. When the calculated similarity is greater than the threshold, a matching pair of feature points is generated. 8. The paper pattern recognition method according to claim 6, characterized in that the feature points are SURF or SIFT feature points. 9. The paper pattern identification method according to claim 1, characterized in that, The step S3-1 includes: using the MSAC algorithm to eliminate erroneous feature point matching pairs among the feature point matching pairs obtained in the step S2, using the reference paper texture image as a standard, and based on the remaining valid feature point matching pairs, Estimate the affine transformation matrix from the paper texture image to be identified to the reference paper texture image; The step S3-2 includes: using the affine transformation matrix to perform affine transformation on the paper texture image to be identified; The step S3-3 includes: stacking the transformed paper texture images, selecting a threshold S, and comparing the area of the overlapping portion after the transformation to the total area of the reference paper texture image or the paper texture image to be identified. When it is greater than S, the largest inscribed rectangle of the overlapping part after transformation is intercepted as the area of interest. 10. The paper texture identification method according to claim 1, characterized in that, before step S4, it includes: According to the image size parameters given in advance, the areas of interest of the reference paper texture image and the paper texture image to be identified are adjusted to the same size. 11. The paper texture identification method according to claim 10, characterized in that the step S5 includes: Map the enhanced texture structure of the area of interest of the paper texture image to be identified and the reference paper texture image to 01 digital space respectively, and use Hamming distance to generate a similarity index to measure the reference paper texture image and the paper texture to be identified. Image correspondence to the similarity of the generated bitstream. 12. The paper pattern identification method according to claim 11, characterized in that the enhanced texture structure is mapped to 01 digital space, and a Hamming distance is used to generate a similarity index to measure the difference between the reference paper pattern image and the The similarity of the generated bit stream corresponding to the paper texture image to be recognized includes: Calculate the mean of the final amplitude response matrix and phase angle response matrix of the reference paper texture image and the area of interest of the paper texture image to be identified respectively, compare each number in the response matrix with the mean, and compare the mean The larger one is 1, and the smaller one is 0. After expanding the two digital matrices by row or column, the resulting bit streams are spliced together and used as the reference paper texture image and the paper texture image to be identified respectively. digital paper texture; Calculate the Hamming distance between the reference paper texture image and the digital paper texture of the paper texture image to be identified, and use the proportion of the Hamming distance to the total length of the digital paper texture as the distance between the reference paper texture and the paper texture to be identified. similarity index; Select a threshold t. If the similarity index is greater than the threshold t, the recognition fails. If the similarity index is less than the threshold t, the recognition is successful.