A multi-layer anti-counterfeiting label recognition system and method

Abstract

The present invention relates to the technical field of anti-counterfeiting labels, specifically a multi-layer anti-counterfeiting label recognition system and method, including: an image acquisition module, a feature layering module, an anti-counterfeiting authentication module, a logic verification module, and an anti-counterfeiting evaluation module; by collecting anti-counterfeiting label images and dividing the anti-counterfeiting label images into multiple sub-images according to the connectivity of each position on the image; performing feature layering processing on the sub-images, extracting image features where the image feature network is associated with the physical layer and the logical layer, and obtaining the layer probabilities and verification methods of the sub-images at each level; performing parallel verification and serial verification on the image features in sequence to form a logical verification relationship diagram of the sub-images; searching for the optimal decision path of the logical verification relationship diagram at each level; taking the optimal decision path of the logical verification relationship diagram as a constraint condition, merging and analyzing each constraint condition to obtain a multi-layer anti-counterfeiting verification network; achieving the accuracy and efficiency of anti-counterfeiting label recognition.

Description

Technical Field

[0001] The present invention relates to the technical field of anti-counterfeiting labels, and specifically to a multi-layer anti-counterfeiting label recognition system and method. Background Art

[0002] Currently, the conventional anti-counterfeiting process is to use an electronic device to scan the anti-counterfeiting label on an item, and then receive the source image of the anti-counterfeiting label collected at the factory by the system, and compare whether the scanned anti-counterfeiting label is consistent with the source image to identify authenticity; however, when using an electronic device to identify a large number of anti-counterfeiting labels in different forms, it is difficult to adjust according to the verification method of the currently input anti-counterfeiting label, resulting in a low label verification efficiency under the multi-source integration verification of the anti-counterfeiting label.

[0003] For example, Chinese Patent Publication No. CN116227524A discloses a method for generating and verifying anti-counterfeiting codes and an anti-counterfeiting system based on labels. The method includes: the generation party of the anti-counterfeiting label obtains an initial anti-counterfeiting image containing a graphic code for anti-counterfeiting, and splits the initial anti-counterfeiting image into multiple sub-images; the multiple sub-images are printed on the label substrate in multiple times to generate a printed anti-counterfeiting image corresponding to the graphic code, and there are random offsets between the printing positions of the multiple sub-images on the label substrate; the verification party of the anti-counterfeiting label obtains the anti-counterfeiting image to be verified collected by the user through the terminal; obtains the printed anti-counterfeiting image corresponding to the graphic code included in the anti-counterfeiting image to be verified; and verifies the authenticity of the anti-counterfeiting image to be verified according to the image similarity between the anti-counterfeiting image to be verified and the printed anti-counterfeiting image.

[0004] For example, Chinese Patent Publication No. CN113435219A discloses an anti-counterfeiting detection method, device, electronic device and storage medium. For the target image generated by a consumer scanning an anti-counterfeiting label and the source image collected when the anti-counterfeiting label leaves the factory, it can comprehensively compare the similarity degree of the anti-counterfeiting points of the two in terms of both position coordinates and color, and determine the authenticity of the label scanned by the consumer based on this similarity degree.

[0005] The prior art illustrates that the anti-counterfeiting label is processed by generating random offsets in the position of the anti-counterfeiting label, and relative colors are used to complete the recognition of the anti-counterfeiting label. However, these recognition methods are mostly single methods and cannot perform combined processing according to multiple groups of currently input anti-counterfeiting labels, resulting in a reduction in efficiency in the scenario of a large number of identifications of anti-counterfeiting labels and problems in the management of verification methods. Summary of the Invention

[0006] In order to solve the above technical problems, the technical solution adopted by the present invention is: a multi-layer anti-counterfeiting label recognition system, including: an image acquisition module, configured to acquire an anti-counterfeiting label image and sequentially divide the anti-counterfeiting label image into multiple sub-images according to the connectivity of each position on the image.

[0007] A feature stratification module, which is used to perform feature stratification processing on sub-images to obtain an image feature network; extract the image features in the image feature network that have feature associations with the physical layer and the logical layer, and after setting a hierarchical division standard for the image features, determine the hierarchical probabilities and verification methods of the sub-images at each level.

[0008] An anti-counterfeiting authentication module, which is used to perform parallel verification and serial verification on the image features in turn according to the verification methods of the sub-images at each level to obtain logical verification data, and form a logical verification relationship graph of the sub-images.

[0009] A logical verification module, which is used to obtain multiple candidate paths based on the data at the starting and ending positions of the logical verification relationship graph, and use the maximum path and the minimum path of the logical verification relationship graph as the search interval to search for the optimal decision path of the logical verification relationship graph at each level.

[0010] An anti-counterfeiting evaluation module, which is used to take the optimal decision path of the logical verification relationship graph as a constraint condition, merge and analyze each constraint condition, determine the shortest execution path after the merger of each constraint condition in turn, and combine multiple constraint conditions according to the shortest execution path to obtain a multi-layer anti-counterfeiting verification network.

[0011] A multi-layer anti-counterfeiting label recognition method, including: S1, inputting an anti-counterfeiting label image into an image segmentation model to generate a region prediction map, and determining whether there are connected regions in each region, and dividing to obtain each sub-image corresponding to the anti-counterfeiting label image.

[0012] S2, extracting each image factor from the sub-images, and determining the label structure corresponding to the image factor according to the distribution of each image factor, and forming a mapping relationship of each image factor; setting an image feature network based on the association between the mapping relationship of the image factor and the physical layer features and the logical layer features.

[0013] S3, calculating the hierarchical probabilities in the image feature network, determining the verification methods at each level, and forming a logical verification relationship graph of the sub-images.

[0014] S4, obtaining multiple candidate paths based on the data at the starting and ending positions of the logical verification relationship graph, and using the maximum path and the minimum path of the logical verification relationship graph as the search interval to search for the optimal decision path of the logical verification relationship graph at each level.

[0015] S5, taking the optimal decision path of the logical verification relationship graph as a constraint condition, merging and analyzing each constraint condition, determining the shortest execution path after the merger of each constraint condition in turn, and combining multiple constraint conditions according to the shortest execution path to obtain a multi-layer anti-counterfeiting verification network.

[0016] The beneficial effects of the present invention are as follows: First, after dividing the sub-images according to the connectivity of each position on the anti-counterfeiting label image, the present invention associates the image features on the sub-images with the physical layer and the logical layer to achieve multi-dimensional authentication, which can reduce the misjudgment rate in the identification and processing of anti-counterfeiting labels and reduce the risk of the anti-counterfeiting label being counterfeited by a single feature. Then, the image features are stratified to obtain the hierarchical probabilities and verification methods of each layer of the image feature network, so as to describe the relative situation of each layer after the anti-counterfeiting label is divided at each position. At the same time, the verification methods are combined to facilitate the subsequent comprehensive processing of these verification methods for image processing.

[0017] Second, by verifying the parallel features and serial features in the image features, the present invention can ensure the consistency of the image features at each divided layer, and by processing the logical verification relationship diagram, it can know the relative relationship and logical structure between multiple verification methods used at each layer to adapt to the identification of different label structures and improve the identification accuracy of different forms of anti-counterfeiting labels.

[0018] Third, by processing the optimal decision path of the image features at each deeper layer, the present invention can generate an optimal decision path for different types of image features according to the different input images of the current anti-counterfeiting label. The purpose of this path is to quickly identify multiple groups of anti-counterfeiting labels by subtracting some unimportant features to determine the verification situation of the current anti-counterfeiting label. Then, all the identified situations are summarized to form a multi-layer anti-counterfeiting verification network to represent the results obtained from the current identification of the anti-counterfeiting label image, so as to reduce the situation of misidentification caused by single-mode verification and improve the speed of dynamic verification of anti-counterfeiting labels. BRIEF DESCRIPTION OF THE DRAWINGS

[0019] The present invention will be further described below with reference to the drawings and embodiments.

[0020] Figure 1 It is a schematic framework diagram of a multi-layer anti-counterfeiting label recognition system.

[0021] Figure 2 It is a schematic flow diagram of a feature stratification module of a multi-layer anti-counterfeiting label recognition system.

[0022] Figure 3 It is a schematic flow diagram of an anti-counterfeiting authentication module of a multi-layer anti-counterfeiting label recognition system.

[0023] Figure 4 It is a schematic flow diagram of a logical verification module of a multi-layer anti-counterfeiting label recognition system.

[0024] Figure 5 It is a schematic flow diagram of a multi-layer anti-counterfeiting label recognition method. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0025] Embodiments of the present invention will be described in detail below. The embodiments described below are exemplary and are only used to explain the present invention, and should not be construed as a limitation of the present invention. For those without specific technical or conditions noted in the embodiments, the techniques or conditions described in the literature in this field or according to the product description are followed.

[0026] Refer to Figure 1 , a multi-layer anti-counterfeiting label recognition system, comprising: an image acquisition module, a feature stratification module, an anti-counterfeiting authentication module, a logic verification module, and an anti-counterfeiting evaluation module; wherein, the output end of the image acquisition module is connected to the feature stratification module, the output end of the feature stratification module is connected to the anti-counterfeiting authentication module, the output end of the anti-counterfeiting authentication module is connected to the logic verification module, and the output end of the logic verification module is connected to the anti-counterfeiting evaluation module.

[0027] The image acquisition module is used to acquire the anti-counterfeiting label image and sequentially divide the anti-counterfeiting label image into a plurality of sub-images according to the connectivity of each position on the image.

[0028] The feature stratification module is used to perform feature stratification processing on the sub-images to obtain an image feature network; extract the image features in the image feature network that have feature associations with the physical layer and the logical layer, and after setting a hierarchical division standard for the image features, determine the hierarchical probabilities and verification methods of the sub-images at each level.

[0029] The anti-counterfeiting authentication module is used to sequentially perform parallel verification and serial verification on the image features according to the verification methods of the sub-images at each level to obtain logical verification data and form a logical verification relationship graph of the sub-images.

[0030] The logic verification module is used to obtain a plurality of candidate paths based on the data at the starting point and the ending point positions of the logical verification relationship graph, and search for the optimal decision path of the logical verification relationship graph at each level with the maximum path and the minimum path of the logical verification relationship graph as the search interval.

[0031] The anti-counterfeiting evaluation module is used to take the optimal decision path of the logical verification relationship graph as a constraint condition, merge and analyze each constraint condition, sequentially determine the shortest execution path after the merger of each constraint condition, and combine the multiple constraint conditions according to the shortest execution path to obtain a multi-layer anti-counterfeiting verification network.

[0032] The physical layer includes but is not limited to the following: Anti-tear layer: special glue is used to reveal hidden text after peeling off to prevent label transfer. Optical variable layer: laser engraving and optically variable ink technology are used to present dynamic colors or patterns at different angles, such as laser rain patterns and laser effects. Material composite layer: embedded with security thread paper, anti-counterfeiting film or micro-nano structure to increase the difficulty of physical imitation. RFID / NFC chip: stores unique encrypted ID and supports contactless reading, such as mobile phone NFC verification. Dynamic QR code: uses "one object, one code" technology, and displays real-time verification results and traceability information after scanning the code.

[0033] The logic layer includes but is not limited to the following: AES / RSA encryption: Encrypt label data (such as production batch, serial number) to prevent data tampering. Dynamic encoding: Make label information unpredictable through variable encoding technology such as timestamp + random number. Frequency domain watermark: Embed the brand logo into the Fourier transform domain of the label image, which is invisible to the naked eye but can be extracted through algorithms. Fluorescent / thermosensitive ink: Display hidden information under specific lighting, such as ultraviolet light or temperature.

[0034] Different anti-counterfeiting labels will have different image features in their actual physical form to identify the content, and there will also be different logical connections between the physical and logical levels. For example, the part of the ink that changes color due to different viewing angles can be identified, and the color temperature and other contents of this color change can be associated with the physical level. The verification code, serial number, etc. can be used as the logical level to verify the anti-counterfeiting label. The anti-counterfeiting label can also be described according to the content that needs to be displayed by adding water in the anti-counterfeiting label as its logical level content, and other image features as the physical level. It can be clearly found here that the content set in the same position or similar position may indicate that the content represents the verification results of different levels. At this time, multi-level verification and identification are required to identify whether the current label is normal. At the same time, the behavioral data of different labels during verification and identification can be used to describe whether the current execution process is a corresponding counterfeiting method. As shown in Table 1, whether the current anti-counterfeiting label has related problems after multi-layer verification.

[0035] Table 1. Anti-counterfeiting label identification diagram

[0036]

[0037] Table 1 illustrates the intent and related situations of the label recognition in the case of partial label recognition. These data will be sent to the data verification terminal when the user performs label recognition. Here, the data verification terminal will comprehensively judge the uploaded image to determine whether the label is real and whether there are corresponding problems. When setting up a multi-layer anti-counterfeiting verification network subsequently, the verification results of the current anti-counterfeiting label image under the corresponding verification method will be mainly identified in the form of a network diagram to determine whether there will be situations related to the problems of the anti-counterfeiting label when the current anti-counterfeiting label is recognized. Ultimately, external staff can discover whether there are corresponding problems in the currently recognized anti-counterfeiting label by viewing the network generated from this recognized anti-counterfeiting label image.

[0038] When dividing the anti-counterfeiting label image into multiple sub-images, according to the image situation described in each position, the anti-counterfeiting label image is divided into multiple sub-images, and each sub-image contains a part of its anti-counterfeiting label image, such as containing one of logo, verification code, pattern, etc.

[0039] At this time, the anti-counterfeiting label image can be located based on the YOLOv5 object detection model. Then, after grid segmentation of the anti-counterfeiting label image, a coordinate system mapping table is generated. The coordinate system mapping table contains the positions of multiple sub-images, such as the sub-images corresponding to positions like the positioning identification area and the encrypted data area.

[0040] The implementation method of the image acquisition module includes: inputting the anti-counterfeiting label image into the image segmentation model to obtain the region prediction map corresponding to the anti-counterfeiting label area. According to the positions of each region prediction map, judge whether there are connected regions in each region prediction map. If there are connected regions, regard each connected region as a sub-image according to the position of the connected region, and regard the region prediction map except the connected region as an additionally output sub-image; if there are no connected regions, directly output each region prediction map as a sub-image.

[0041] The anti-counterfeiting label area indicates the anti-counterfeiting type represented by this area, such as positions like the positioning identification area and the encrypted data area, or the content such as logo, verification code, pattern, etc. at this position, so as to divide the anti-counterfeiting label image into multiple different-style sub-images. These sub-images will contain multiple bound recognition features according to the different displayed contents during recognition.

[0042] Connected regions may be an important feature in anti-counterfeiting label design. By checking the presence, shape, size and other attributes of these regions, it can help verify the authenticity of the product. For example, a genuine anti-counterfeiting label may contain some complex patterns or structures that are difficult to copy, and these features can be identified by detecting connected regions. Connected regions can also be used to evaluate whether the anti-counterfeiting label is intact. If a region that should be connected shows breaks or missing parts, it may indicate that the label has been tampered with or is not original.

[0043] The connected regions in the anti-counterfeiting label usually correspond to key anti-counterfeiting features, and their contents can be classified into the following categories: microstructures, structural textures, logical anti-counterfeiting features, positioning identification areas, holographic pattern areas, and fluorescent ink areas.

[0044] After recognizing the corresponding situation, these connected regions will divide these regions into multiple sub-images, and verify each image separately according to the verification methods required by these images to determine the relative situation of the current anti-counterfeiting label.

[0045] Microstructures are generally represented as microtext / patterns, which are usually tiny characters invisible to the naked eye, such as 0.1mm characters, and require a microscope or high magnification to be recognized; these features will collect the microstructures existing on the current anti-counterfeiting label through a corresponding imaging device, and judge whether they are legal microstructures according to the shape and edge sharpness of the connected regions.

[0046] Structural textures are represented as complex textures formed by randomly distributed fibers, optically variable particles, etc., and identify the current structural texture features by analyzing the shape complexity and distribution density of the connected regions.

[0047] Logical anti-counterfeiting features are generally encrypted data areas, including connected regions containing encrypted QR codes or barcodes, which store the unique ID or hash value of the product; after decoding and verifying the digital signature, such as the SM2 / SM3 national cryptographic algorithm, the corresponding logical anti-counterfeiting features can be obtained.

[0048] The positioning identification area contains special geometric figures, such as concentric circles and crosshairs, for coordinate system alignment, and determines the content of this identification area by verifying the consistency of geometric attributes such as the center coordinates and line segment angles of the connected regions.

[0049] The holographic pattern area contains dynamic optically variable patterns, such as different texts displayed from different perspectives; this part will analyze the color change law of the connected regions when shooting from multiple angles; judge the features obtained by this sub-image through multiple images.

[0050] The fluorescent ink area represents a hidden pattern developed under ultraviolet light, mainly detecting the brightness distribution of the connected regions under a light source with a specific wavelength; this part can also be represented by the color temperature difference under characteristic light, and finally explain whether the sub-image of this part is displayed normally.

[0051] After that, feature layering is performed on the output sub-images to determine the image features for verifying each sub-image, and explain the recognition situation of the anti-counterfeiting label according to the differences between each image feature and the preset image.

[0052] In an embodiment of the present invention, as Figure 2As shown, the implementation method of the feature layering module includes: extracting each image factor from the sub-image, determining the label structure corresponding to each image factor in sequence according to the distribution of each image factor, and forming the mapping relationship of each image factor according to the label structure.

[0053] Associate the mapping relationship of each image factor with the physical layer features and logical layer features, and select the image features corresponding to each image factor.

[0054] Use the image features to construct an image feature network, determine the hierarchical probability of each level of image features under the mapping relationship of the label structure, and set the hierarchical division standard of the image features based on each hierarchical probability.

[0055] When constructing the image feature network with the image features, its implementation method also includes: taking the image factor type of each image feature as the standard, using each image feature as a node, taking the Pearson correlation coefficient between each image feature as the weight, connecting each image feature to obtain an image feature network, calculating the conditional probability distribution of each level of image features in the image feature network, setting the probability threshold of each level according to the conditional probability distribution of each level of image features, and obtaining the hierarchical structure of the current image feature division. Here, the conditional probability is the above-mentioned hierarchical probability. Finally, the divided image feature network will be presented as a binary tree, etc., with multiple levels divided by the probability threshold of each level, and the details of the conditional probability described by each level, such as the conditional probability when the physical layer recognition is effective and the high-frequency component ratio is large. These contents will represent the probability value of the current anti-counterfeiting label under different recognition situations and whether the recognition method is effective; the Pearson correlation coefficient between the image features will be calculated using the data of the extracted image features after preprocessing to eliminate the dimension to determine whether there is an association between multiple image features.

[0056] Each image factor described here represents the distribution position of the corresponding connected region in the sub-image. After representing the images corresponding to each image factor with labels and explaining the content corresponding to these images, if there is a clear corresponding relationship between a certain image factor and the label structure, the mapping relationship is regarded as a direct mapping, that is, a specific pattern in the anti-counterfeiting label corresponding to a certain factor can directly extract the corresponding feature. For example, if this image factor represents the content of the positioning identification area, this content can be directly recognized through a single image, then there is a direct mapping relationship at this time. If the image factor represents the content on the holographic pattern area and multiple image factors need to be considered comprehensively, then an indirect mapping relationship needs to be set; then these mapping relationships are associated with the required physical layer features and logical layer features to obtain the image features of each layer. For example, in the form of Table 2, after classifying the image factors, various sub-images existing in the current anti-counterfeiting label image can be recognized.

[0057] Table 2. Definition of Image Factors

[0058]

[0059] Some of the describable image factor types are illustrated in Table 2. These image factors will be identified using corresponding technologies to determine how many different forms of sub-images each anti-counterfeiting label should be divided into.

[0060] For example, after the mapping relationship is constructed, if 3 approximately circular contours are detected → mapped to the positioning area; if a square modular distribution is detected → mapped to the QR code in the data area, and it is monitored that the proportion of high-frequency components > 25% → mapped to the microtext area. For other identified contents, they will be mapped according to the corresponding types of their image factors to determine the mapping relationships of each image factor, as shown in Table 3 and Table 4. When there are images related to microtext and fluorescent fibers, their association methods can be illustrated as shown in the following table.

[0061] Table 3. Association of Physical Layer Features

[0062]

[0063] Table 4. Association of Logical Layer Features

[0064]

[0065] Table 3 and Table 4 mainly illustrate how each image factor can be associated. After that, what main verification features and verification contents are required for these associated contents, and finally a hierarchical network is formed. Then, the conditional probabilities of each layer are calculated to obtain the corresponding layer probabilities. As shown in Table 5, when verifying the image features of each layer, the corresponding image factors will be combined, and after calculating the relative layer probability, it is compared whether this value is greater than the probability threshold to divide multiple recognition levels to determine which method to verify the authenticity of the anti-counterfeiting label.

[0066] Table 5. Hierarchical Division Criteria

[0067]

[0068] Under the hierarchical division corresponding to Table 5, if the confidence level of the coding factor detected at the L1 level is 0.7 → trigger L1 verification, the energy value of the texture factor is 0.28 (within [0.12, 0.40]), and it is found that the high-frequency proportion is 24%; at the L2 level, the conditional probabilities of the texture factor and the spectrum factor rise to 0.88, and the conditional probability is greater than 0.85, start microscopic verification. Finally, the combined conditional probability in L3 is 0.93, which is greater than the probability of entering the L3 level; after that, using the verification actions corresponding to each level are regarded as the verification methods of subsequent sub-images at each level.

[0069] That is, the implementation method for determining the verification method of sub-images at each level further includes: extracting the label structure of image features at each level, based on the label structure, using the image factor types described by the image features for joint retrieval, comparing the retrieved results with the image features at each level, and selecting the largest data set after comparison as the verification method of the sub-image at each level.

[0070] Here, it is to perform secondary verification on the image features under each level division to determine whether the retrieved data can be the same as the image features of the current division. When they are the same, the verification method retrieved is used as the verification method of the sub-image in the current anti-counterfeiting label. The label structure will represent the area where its image features are located at each level, such as the positioning area, data area, background area, microtext area, etc., to represent the relative description of the image features.

[0071] In an embodiment of the present invention, as Figure 3 shown, the implementation method of the anti-counterfeiting authentication module further includes: applying a proportional factor to the hierarchical probabilities of each image feature to obtain the adjusted probability thresholds of each level of the image features, setting the hierarchical relative position information of each image feature according to the feature area corresponding to the adjusted probability threshold; the hierarchical relative position information represents the area in the current sub-image containing the corresponding image feature divided by the adjusted probability threshold, and then records the center point coordinates of this feature area, the distance and direction between this area and other areas, and the dependence relationship between this image feature and other image features, and takes these contents as the hierarchical relative position information of each image feature.

[0072] For example, the above proportional factor will set different weights according to different levels of each image feature. For example, taking the probability thresholds on the above-mentioned L1, L2, and L3 layers as an illustrative explanation, at this time, the weights of each level can be adjusted. For example, the probability of L1 is 0.70, and after adjustment: 0.70×1.2 = 0.84; the probability of L2 is 0.85, and after adjustment: 0.85×1.0 = 0.85; the probability of L3 is 0.90, and after adjustment: 0.90×0.8 = 0.72; this shows that the currently used probability threshold will be adjusted according to each image feature under normal recognition conditions. At this time, the threshold of the first layer is increased and the threshold of the third layer is decreased to determine whether the divided image features will change after the current method is adjusted, and the information with these changes is used as the main content for subsequent processing.

[0073] Using the hierarchical relative position information of each image feature, parallel verification and serial verification are performed on the image features in sequence to determine the independent features and dependent features existing in the image features, and the verification results of the independent features and dependent features are regarded as logical verification data.

[0074] Parallel features mainly verify whether image features are independent features, while serial features verify whether image features are dependent features; independent features mean that the recognition result of a single image can represent the recognition result of the current anti-counterfeiting label in a certain aspect. When there is a problem with this recognition, it can be directly considered that there is a problem with the anti-counterfeiting label, or that the anti-counterfeiting label is fake; dependent features mean that it is necessary to combine and comprehensively judge the corresponding contents of multiple image features to determine the problem with the current anti-counterfeiting label, or to use this to determine whether the anti-counterfeiting label is authentic.

[0075] Simultaneous parallel processing utilizes multi-threaded or distributed computing resources to verify each independent feature at the same time, which can greatly improve efficiency, especially when processing high-resolution images; if any independent feature fails to pass the verification, the anti-counterfeiting label is immediately marked as suspicious.

[0076] Serial processing tends to verify in sequence, and the next step is performed only when each dependent feature is successful. Only when all dependent features are verified can the anti-counterfeiting label be preliminarily considered to be authentic; if there is any discrepancy in the middle, further investigation is required. By dividing the anti-counterfeiting labels into these two forms, the image data input by the current anti-counterfeiting label can be adjusted to achieve the common processing of multiple different types of anti-counterfeiting labels as much as possible.

[0077] Subsequently, the implementation methods of performing parallel verification and serial verification on the image features in sequence include: verifying the fields of each node in the image feature network, determining the basic verification data, and configuring the basic verification data with an asynchronous message queue mechanism; the field verification is to describe whether the text description used for the verification method and the image features is normal, and the content of this part of the text description is executed using an asynchronous message queue mechanism to determine whether the verification of the current anti-counterfeiting label is in the synchronous processing scope or the asynchronous processing scope.

[0078] If the asynchronous message queue mechanism is configured, and the node in the current image feature network points to the next node, the image feature corresponding to the current node is regarded as a dependent feature, otherwise it is regarded as an independent feature; at this time, it is determined whether the current asynchronously executed node in the divided image feature network depends on the execution results of other nodes. If it is dependent, it means that sequential inspection is required, otherwise it can be implemented asynchronously.

[0079] After the cross-execution of the independent features and dependent features, the verification methods of each level are compared to see whether they are consistent with the independent features and dependent features at the intersection position. If they are consistent, the data corresponding to the image features are output as logical verification data.

[0080] If they are inconsistent, the hierarchical relative position information of the independent features and the dependent features is recorded, and the independent features and the dependent features are mapped to each other according to the positions recorded in the hierarchical relative position information, and output as logic verification data.

[0081] At this time, it is required that after the independent feature and the dependent feature are executed, whether the results of the independent verification and the dependent verification can be kept consistent at the intersection position. If they can be kept consistent, the data for verification will be output. When they are inconsistent, these data will be marked in a mutually mapped manner and then output as logical verification data. After that, the output data will be composed into a logical verification relationship graph, which mainly records these relatively independent and mutually dependent image features and connects the logical relationships existing between these features.

[0082] According to the cross-execution order of the independent feature and the dependent feature in the logical verification data, each node of the image feature network corresponding to the independent feature and the dependent feature is connected to form a logical verification relationship graph of the sub-image.

[0083] At this time, the connection is mainly carried out according to the features included in the logical verification data, and then according to the cross-execution order of these features under the asynchronous mechanism, the data corresponding to these two features are connected to obtain a logical graph about the current execution order. After that, this logical graph is used to complete the processing of the subsequent decision path.

[0084] In an embodiment of the present invention, a search mechanism is mainly introduced in the logical verification module to determine the relative dependence relationship and possible paths between the verification methods of each image feature in the logical verification relationship graph; for example, depth-first search (DFS) or breadth-first search (BFS) is used for traversal to obtain the paths processed by these verification methods and image features.

[0085] After that, the verification status of each node in the logical verification relationship network is judged, such as the verification status of passed, not passed or feasible and unfeasible, plus the confidence score or other indicators of each node, to calculate a comprehensive weight for each path; in this way, a path with the highest weight and a path with the lowest weight are obtained, and these two paths represent the maximum path and the minimum path. Then, a search interval is obtained according to the weights of the maximum path and the minimum path, and an optimal decision path is found for the paths existing within these intervals.

[0086] As Figure 4 shown, the implementation manner of the logical verification module includes: identifying the image features of each path node on the candidate path, and determining that the image features include at least one of a shape factor, a texture factor, a spectrum factor, and a coding factor.

[0087] Based on the types of image features included in each path node, identify the maximum path and the minimum path among the candidate paths in terms of the number of different types of image features they contain. At this time, the processing methods of the candidate paths under the need to identify different image features are identified, and among these processing methods, the path with the largest weight and the path with the smallest weight are identified. Then, multiple maximum paths and minimum paths are obtained according to the different types of image features they contain.

[0088] Use the number of types of image features corresponding to the maximum weight value in the maximum path and the minimum path as the search range, retrieve each candidate path, extract the shortest path connecting the path nodes in each candidate path, and obtain the optimal decision path. At this time, use the shortest path algorithm for each path node in the candidate paths within the search range to obtain a shortest path from the starting point to other points, and sequentially obtain the optimal decision path to be output.

[0089] For example, at this time, through four factors corresponding to the image features, identify the optimal decision path in a group of candidate paths. Among them, the weight of candidate path 1 is the largest, and the weight of candidate path 2 is the smallest. At this time, the number of types of image features of candidate path 1 is 3, and the number of types of image features of candidate path 2 is 2. At this time, mainly search for candidate paths containing the number of types of image features from 2 to 3, and then find the shortest path at this time. This usage method can retain as many features as possible, can retain the complete topological structure, and can retain key features while avoiding redundant features; at the same time, it can reduce the problem of incorrect identification of anti-counterfeiting labels caused by feature loss, and finally improve the overall identification efficiency and speed.

[0090] In an embodiment of the present invention, when the constraint conditions are merged, the content corresponding to the optimal decision path is merged, and the merged content is compared to see if it can be used for anti-counterfeiting label verification after being combined with the verification method. If it can be verified, the shortest execution path of each merged constraint condition is output. At this time, the output shortest execution path represents the minimum number of data nodes required to implement the corresponding verification method or image feature processing after satisfying the data represented by each constraint condition, and it is the node that can complete the verification of this method. Then, these shortest execution paths are combined using the constraint conditions to obtain a multi-layer anti-counterfeiting verification network that can meet the current anti-counterfeiting label identification.

[0091] At the same time, when generating the logical verification relationship diagram, if multiple different anti-counterfeiting label images are input, then multiple logical verification relationship diagrams will be generated at this time, and there will also be multiple groups of data for the optimal decision path. Merging these data can reduce the processing efficiency in the case of automatic identification and processing of multiple anti-counterfeiting labels, and reduce the problem of incorrect identification caused by partial label damage.

[0092] Therefore, the implementation method of the anti-counterfeiting evaluation module includes: grouping using the logical relationships of each constraint condition, constructing a constraint relationship matrix, calculating the similarity of each element in the constraint relationship matrix, and merging the elements with a similarity greater than the similarity threshold; at this time, the constraint relationship matrix will contain the path nodes in the optimal decision path, each element is represented as the path node of the optimal decision path, and then these nodes are grouped, and then these nodes are compared for similarity and identity to complete the merging. When calculating the similarity, the Pearson correlation coefficient is used for calculation, and the path nodes with a similarity greater than 0.7 are merged to achieve the merging of each element in the constraint relationship matrix.

[0093] Connect the elements in the merged constraint relationship matrix, perform path retrieval based on the path length and verification cost of the corresponding constraint conditions of each element, obtain the shortest execution path, and use the shortest execution path to summarize each constraint condition to combine and obtain a multi-layer anti-counterfeiting verification network.

[0094] When performing path retrieval, the purpose is to combine multiple constraint conditions to find a convenient implementation path after combination. These paths will represent a path for summarizing the decision-making content after multiple optimal decision paths make decisions based on different label images. Then, according to this path, the paths of multiple decisions can be combined to understand the processing process under multi-type label recognition.

[0095] When connecting the elements in the merged constraint relationship matrix, use the path length in the corresponding constraint condition of the element as the weight of the element, and use the verification cost of the two connected elements as the weight of the edge. This path length will represent the number of nodes in the optimal decision path represented by the constraint condition, and the verification cost will use the time required for recognition processing to complete the image features corresponding to the current element as the verification cost. Then, obtain the shortest path that these elements can complete the connection, and this path is regarded as the shortest execution path output. Finally, use the elements corresponding to the shortest execution path to combine multiple optimal decision paths to complete the multi-layer anti-counterfeiting verification network for the current anti-counterfeiting label recognition. The multi-layer anti-counterfeiting verification network will represent the models required for the recognition of various anti-counterfeiting labels to improve the integration processing of various anti-counterfeiting labels when dividing sub-images of anti-counterfeiting labels or inputting various forms of anti-counterfeiting labels.

[0096] As Figure 5 shown, the present invention also provides a multi-layer anti-counterfeiting label recognition method, including: S1, inputting the anti-counterfeiting label image into an image segmentation model to generate a region prediction map, and determining whether there are connected regions in each region, and dividing to obtain each sub-image corresponding to the anti-counterfeiting label image.

[0097] S2. Extract each image factor from the sub-image, determine the label structure corresponding to the image factor according to the distribution of each image factor, and form the mapping relationship of each image factor; set the image feature network based on the association between the mapping relationship of the image factor and the physical layer feature and the logical layer feature.

[0098] S3. Calculate the probabilities of each layer in the image feature network, determine the verification methods of each layer, and form the logical verification relationship diagram of the sub-image.

[0099] S4. Obtain multiple candidate paths based on the data at the starting and ending positions of the logical verification relationship diagram, and search for the optimal decision path of the logical verification relationship diagram at each layer with the maximum path and the minimum path of the logical verification relationship diagram as the search interval.

[0100] S5. Taking the optimal decision path of the logical verification relationship diagram as a constraint condition, merge and analyze each constraint condition, sequentially determine the shortest execution path after the merger of each constraint condition, and combine multiple constraint conditions according to the shortest execution path to obtain a multi-layer anti-counterfeiting verification network.

[0101] Although the embodiments of the present invention have been shown and described above, it can be understood that the above embodiments are exemplary and should not be construed as limiting the present invention. Those of ordinary skill in the art can make changes, modifications, substitutions, and variations to the above embodiments within the scope of the present invention, and still be covered by the protection scope of the present invention.

Claims

1. A multi-layer anti-counterfeiting label identification system, characterized in that: include: An image acquisition module is used to acquire an anti-counterfeiting label image and divide the anti-counterfeiting label image into a plurality of sub-images in sequence according to the connectivity of each position on the image; The feature stratification module is used to perform feature stratification processing on the sub-images to obtain an image feature network; extract image features that are feature-associated with the physical layer and the logical layer in the image feature network, set a hierarchical division standard for the image features, and determine the hierarchical probability and verification method of the sub-images at each level; The anti-counterfeiting authentication module is used to perform parallel verification and serial verification on the image features in turn according to the verification methods of the sub-images at each level, obtain logical verification data, and form a logical verification relationship diagram of the sub-images; A logic verification module is used to obtain multiple candidate paths based on the data of the starting point and end point positions of the logic verification relationship graph, and to search for the optimal decision path at each level of the logic verification relationship graph using the maximum path and the minimum path of the logic verification relationship graph as the search interval; The anti-counterfeiting evaluation module is used to combine and analyze the constraints with the optimal decision path of the logic verification relationship diagram as the constraint condition, determine the shortest execution path after the constraints are combined in turn, and combine multiple constraints according to the shortest execution path to obtain a multi-layer anti-counterfeiting verification network.

2. A multi-layer anti-counterfeiting label identification system according to claim 1, characterized in that: The implementation methods of the image acquisition module include: The anti-counterfeiting label image is input into the image segmentation model to obtain the regional prediction map corresponding to the anti-counterfeiting label area. According to the position of each regional prediction map, it is determined whether there is a connected area in each regional prediction map. If there is a connected area, each connected area is regarded as a sub-image according to the position of the connected area, and the regional prediction maps except the connected area are used as additional output sub-images; if there is no connected area, each regional prediction map is directly output as a sub-image.

3. A multi-layer anti-counterfeiting label identification system according to claim 1, characterized in that: The implementation methods of the feature layering module include: Extract each image factor from the sub-image, determine the label structure corresponding to each image factor in turn according to the distribution of each image factor, and form a mapping relationship of each image factor according to the label structure; Associating the mapping relationship of each image factor with the physical layer feature and the logical layer feature, and selecting the image feature corresponding to each image factor; Image feature networks are constructed using image features to determine the hierarchical probabilities of image features at each level under the mapping relationship of the label structure, and based on the probabilities of each level, hierarchical division criteria for image features are set.

4. A multi-layer anti-counterfeiting label identification system according to claim 3, characterized in that: When the image feature is used to construct the image feature network, the implementation method also includes: Taking the image factor type of each image feature as the standard, each image feature is taken as a node, and the Pearson correlation coefficient between each image feature is used as the weight, each image feature is connected to obtain an image feature network, and the conditional probability distribution of image features at each level of the image feature network is calculated. According to the conditional probability distribution of image features at each level, the probability threshold of each level is set to obtain the hierarchical structure of the current image feature division.

5. A multi-layer anti-counterfeiting label identification system according to claim 3, characterized in that: The implementation method of determining the verification method of the sub-image at each level also includes: The label structure of image features at each level is extracted. Based on the label structure, the image factor type described by the image features is used for joint retrieval. The retrieval results are compared with the image features at each level. The largest data set after comparison is selected as the verification method for sub-images at each level.

6. A multi-layer anti-counterfeiting label identification system according to claim 1, characterized in that: The implementation of the anti-counterfeiting authentication module also includes: Applying a proportional factor to the hierarchical probability of each image feature to obtain an adjusted probability threshold of each level of the image feature, and setting the hierarchical relative position information of each image feature according to the feature area corresponding to the adjusted probability threshold; Using the hierarchical relative position information of each image feature, the image features are sequentially verified in parallel and serially, the independent features and dependent features existing in the image features are determined, and the verification results of the independent features and dependent features are regarded as logical verification data; According to the cross-execution order of independent features and dependent features in the logic verification data, the nodes of the image feature network corresponding to the independent features and the dependent features are connected and combined into a logic verification relationship diagram of the sub-image.

7. A multi-layer anti-counterfeiting label identification system according to claim 6, characterized in that: The implementation methods of performing parallel verification and serial verification on image features in sequence include: Verify the fields of each node in the image feature network, determine the basic verification data, and configure the asynchronous message queue mechanism for the basic verification data; If the asynchronous message queue mechanism is configured, and the node in the current image feature network points to the next node, the image feature corresponding to the current node is regarded as a dependent feature, otherwise it is regarded as an independent feature; After the cross-execution of independent features and dependent features, the verification methods of each level are compared to see whether they are consistent with the independent features and dependent features at the intersection position. If they are consistent, the data corresponding to the image features are output as logical verification data; If they are inconsistent, the hierarchical relative position information of the independent features and the dependent features is recorded, and the independent features and the dependent features are mapped to each other according to the positions recorded in the hierarchical relative position information, and output as logic verification data.

8. The multi-layer anti-counterfeiting label identification system according to claim 1, characterized in that: The implementation of the logic verification module includes: Identify image features of each path node on the candidate path, and determine that the image features include at least one of a shape factor, a texture factor, a spectrum factor, and a coding factor; Based on the types of image features contained in each path node, identify the maximum and minimum paths containing different types of image features in the candidate paths; The number of image feature types corresponding to the maximum weight value in the maximum path and the minimum path is used as the search interval, each candidate path is retrieved, and the shortest path connecting the path nodes in each candidate path is extracted to obtain the optimal decision path.

9. A multi-layer anti-counterfeiting label identification system according to claim 1, characterized in that: The implementation methods of the anti-counterfeiting assessment module include: Use the logical relationship of each constraint condition to group, construct a constraint relationship matrix, calculate the similarity of each element in the constraint relationship matrix, and merge the elements that are greater than the similarity threshold; The elements in the merged constraint relationship matrix are connected, and the path length and verification cost of the constraint conditions corresponding to each element are used to perform path retrieval to obtain the shortest execution path. The constraints are summarized using the shortest execution path to obtain a multi-layer anti-counterfeiting verification network.

10. A multi-layer anti-counterfeiting label identification method, characterized in that: include: S1, input the anti-counterfeiting label image into the image segmentation model, generate a region prediction map, and determine whether there are connected regions in each region, and divide it to obtain sub-images corresponding to the anti-counterfeiting label image; S2, extracting each image factor from the sub-image, and determining the label structure corresponding to the image factor according to the distribution of each image factor, and forming a mapping relationship between each image factor; setting an image feature network based on the association between the mapping relationship of the image factor and the physical layer features and the logical layer features; S3, calculate the probability of each level in the image feature network, determine the verification method of each level, and form a logical verification relationship diagram of the sub-image; S4, obtaining multiple candidate paths based on the data of the starting point and the end point of the logic verification relationship graph, using the maximum path and the minimum path of the logic verification relationship graph as the search interval, and searching for the optimal decision path of the logic verification relationship graph at each level; S5, taking the optimal decision path of the logic verification relationship diagram as the constraint condition, merging and analyzing each constraint condition, determining the shortest execution path after merging each constraint condition in turn, and combining multiple constraint conditions according to the shortest execution path to obtain a multi-layer anti-counterfeiting verification network.

Images