An optical feature driven printed product authentication method, system, medium and product

Abstract

The application discloses a kind of optical feature driven printed matter identification method, system, medium and product, the method of the application includes obtaining the original color set generated by uniform sampling in specified color domain;Original color set is respectively input into the color set prediction model pre-trained under multiple printed matter classification types to obtain the predicted printed color set under the corresponding printed matter classification type, the color set prediction model is pre-trained to establish the mapping relationship between the input original color set, the predicted printed color set output under this printed matter classification type;Color set is used as the optical feature of printed matter, the similarity of the color set of the identified printed matter and the predicted printed color set under each printed matter classification type is used to determine the printed matter classification type corresponding to the identified printed matter.The application aims at focusing on the quantifiable optical feature of pattern in the optical processing process of printing and scanning of printed matter to realize accurate and high-precision printed matter identification classification.

Description

TECHNICAL FIELD

[0001] The present application belongs to the technical field of image processing, and particularly relates to a printed matter identification method and system driven by optical features, a medium and a product. BACKGROUND

[0002] Printed matter identification includes distinguishing whether the printed matter is an original printed matter or a copied printed matter. Existing printed matter identification methods include manual observation method and active protection method based on anti-scanning paper. The manual observation method is the most basic identification means, which includes observing the printing characteristics and paper details of the file by naked eye or with the help of magnifying glass, side light and the like. The details such as characters and lines on the original are usually clear and regular, and the edges of strokes are smooth, while the copied pieces may have some toner particles and screen dots, and the edges of characters and patterns may have jagged or unsmooth phenomena. In addition, the flexibility and strength of the original can be felt by touching the paper, and the flexibility and strength of the original are relatively good and not easy to break, while the copied pieces may feel brittle and prone to breakage. For files with seals or fingerprints, the seal color of the original is generally darker, and the seal color of the copied pieces is brighter. The anti-scanning paper adds special materials or microstructures during papermaking or surface treatment, and when collected by a scanner, it will produce identifiable interference effects. When the paper is used for initial printing, the original output displays normally under naked eye and standard optical conditions, and there is no additional abnormal pattern; and once scanned and reprinted, the optical system of the scanner will interact with the anti-fake features in the paper, resulting in moire fringes, color distortion, local blur or the appearance of preset "VOID" warning marks. By observing these abnormal features that only appear during re-printing, the original printed matter and the re-printed matter can be effectively distinguished. However, the above methods have significant defects and shortcomings in application. The manual observation method depends on the experience and subjective judgment of the operator, and the ability to identify subtle differences is limited, it is difficult to form a unified and quantifiable identification standard, and the efficiency is low, which is not suitable for batch detection scenarios; the anti-scanning paper can provide strong physical protection during production, but it needs to use special materials in advance, and it cannot be used for traceable identification of ordinary paper files that have been circulated, the anti-fake effect may be partially cracked by high-resolution scanning and professional image processing technology, the special coating or microstructure may affect the printing quality and color restoration, and the cost is high, which is not convenient for large-scale application. SUMMARY

[0003] The technical problem to be solved by the present application is to provide a printed matter identification method and system driven by optical features, a medium and a product, which focuses on the quantifiable optical features of patterns in the optical processing process of printing and scanning to realize accurate and high-precision printed matter identification and classification.

[0004] To solve the above technical problems, the technical scheme adopted by the present application is:

[0005] An optical feature driven printed matter identification method, comprising the following steps:

[0006] S101, obtaining an original color set generated by uniform sampling in a specified color domain;

[0007] S102, inputting the original color set into a color set prediction model pre-trained under each printed matter classification type to obtain a predicted printed color set under the corresponding printed matter classification type, wherein the color set prediction model is pre-trained to establish a mapping relationship between the input original color set and the output predicted printed color set under the printed matter classification type;

[0008] S103, using the color set as the optical feature of the printed matter, and determining the corresponding printed matter classification type of the identified printed matter by using the similarity between the color set of the identified printed matter and the predicted printed color set under each printed matter classification type.

[0009] Optionally, in the step S101 of obtaining the original color set generated by uniform sampling in the specified color domain, the specified color domain refers to an RGB color domain, and the obtaining of the original color set generated by uniform sampling in the specified color domain comprises respectively performing equal-interval sampling on N value points in the red channel R, the green channel G and the blue channel B of the RGB color domain, thereby obtaining an original color set composed of N×N×N color different value points.

[0010] Optionally, the step S102 further comprises training the color set prediction model under each printed matter classification type, and the training step of the color set prediction model under any printed matter classification type comprises:

[0011] S201, obtaining a printed matter of the original color set under the current printed matter classification type, wherein the printed matter is obtained by printing or printing the digital color card of the original color set under the current printed matter classification type;

[0012] S202, detecting and obtaining the printed color set of the printed matter by using a colorimeter;

[0013] S203, normalizing the printed color set to obtain a printed color set sample;

[0014] S204, constructing a training data set of the original color set and the color set sample of the corresponding printed matter;

[0015] S205, a color set prediction model under the current printed product classification type is established, an input of the color set prediction model is an original color set, an output is a predicted post-printing color set, the color set prediction model is trained by using the training data set, so that a mapping relationship between the input original color set and the output predicted post-printing color set under the printed product classification type is established.

[0016] Optionally, the similarity between the color set of the identified printed product and the predicted post-printing color set under each printed product classification type is used to determine the printed product classification type corresponding to the identified printed product in step S103, and the method comprises the following steps.

[0017] S301, the color set of the identified printed product is obtained.

[0018] S302, the color set of the identified printed product and the predicted post-printing color set under each printed product classification type are input into the pre-trained printed product identification model as input data, the similarity score of the color set of the identified printed product and the predicted post-printing color set under each printed product classification type is obtained by the printed product identification model, and the printed product classification type with the largest similarity score is determined as the printed product classification type corresponding to the identified printed product.

[0019] Optionally, the printed product identification model in step S103 is a twin network, the twin network comprises a shared feature extractor and a plurality of independent classification heads, the feature extractor is used to extract the color set of the identified printed product or the post-printing color set under a certain printed product classification type to obtain an embedding vector representing the color set feature, the number of classification heads is the same as the number of printed product classification types, and each classification head is used to generate a similarity score between the embedding vectors of the color set of the identified printed product and the post-printing color set under the corresponding printed product classification type.

[0020] Optionally, the feature extractor in the twin network comprises a three-layer fully connected network and a splicing module, the color set of the identified printed product or the post-printing color set under a certain printed product classification type is extracted by the three-layer fully connected network to obtain a 16-dimensional embedding vector representing the color set feature, and the splicing module is used to splice the embedding vectors of the color set of the identified printed product and the post-printing color set under the corresponding printed product classification type to obtain a spliced embedding vector; the classification head in the twin network is composed of a two-layer fully connected network, which is used to map the spliced embedding vector to a similarity score; and a total loss function used by the twin network during training is a weighted sum of a contrast loss and a binary cross entropy loss.

[0021] Optionally, the multiple printed product classification types in step S102 include original printed products and copied printed products, or the multiple printed product classification types in step S102 include qualified original printed products and unqualified original printed products; the original printed product refers to a printed product printed or printed for the first time according to a design drawing, and the copied printed product refers to a printed product obtained by printing or printing again after scanning based on an original printed product or a copy of the original printed product.

[0022] The application also provides an optical feature driven printed product identification system, comprising a microprocessor and a memory connected to each other, wherein the microprocessor is programmed or configured to execute the optical feature driven printed product identification method.

[0023] The application also provides a computer readable storage medium, wherein a computer program or instructions are stored in the computer readable storage medium, and the computer program or instructions are programmed or configured to execute the optical feature driven printed product identification method by a processor.

[0024] The application also provides a computer program product, comprising a computer program or instructions, which are programmed or configured to execute the optical feature driven printed product identification method by a processor.

[0025] Compared with the prior art, the application mainly has the following beneficial effects:

[0026] 1. The method comprises obtaining an original color set uniformly sampled in a specified color domain; inputting the original color set into a color set prediction model pre-trained under multiple printed product classification types to obtain a predicted printed color set under the corresponding printed product classification type, and the color set prediction model is pre-trained to establish a mapping relationship between the input original color set and the output predicted printed color set under the printed product classification type; using the color set as the optical feature of the printed product, determining the corresponding printed product classification type of the identified printed product by using the similarity between the color set of the identified printed product and the predicted printed color set under each printed product classification type, thereby focusing on the quantifiable optical feature (color set) of the pattern in the printing and scanning optical processing process of the printed product. Due to the optical characteristics of the scanner and the printer, the color space conversion and the influence of the physical medium, the scanning and reprinting process will introduce specific color distortion modes. By collecting the RGB color features and establishing a color set prediction model (such as SVR), these color difference modes can be quantified, so that the similarity between the color set of the identified printed product and the predicted printed color set under each printed product classification type can further learn the discriminative representation of the printing and scanning and reprinting samples in the feature space, thereby realizing accurate and high-precision printed product identification and classification.

[0027] 2. To address the difficulty of quantifying subtle color differences during printing and scanning-to-reprinting using traditional manual identification, this invention inputs the original color set into pre-trained color set prediction models for various printing product classification types to obtain the predicted post-printing color set for the corresponding printing product classification type. The pre-trained color set prediction model establishes a mapping relationship between the input original color set and the output predicted post-printing color set for that printing product classification type, which can accurately quantify subtle color differences during printing and scanning-to-reprinting, and achieve precise quantitative modeling of color distortion patterns under different printing methods.

[0028] 3. This invention can support the identification and classification of various printed materials, including original printed materials and reproduced printed materials, or qualified original printed materials and unqualified original printed materials, etc., and has the advantage of wide applicability. Attached Figure Description

[0029] Figure 1 This is a schematic diagram of the basic process of the method in Embodiment 1 of the present invention.

[0030] Figure 2 This is a schematic diagram of the training data for the color set prediction model in Embodiment 1 of the present invention.

[0031] Figure 3 This is a schematic diagram of the prediction data of the color set prediction model in Embodiment 1 of the present invention.

[0032] Figure 4 This is a schematic diagram illustrating the principle of printing type classification identification in Embodiment 1 of the present invention. Detailed Implementation

[0033] To enable those skilled in the art to better understand the technical solutions of the present invention, the technical solutions of the present invention will be further described in detail below with reference to the accompanying drawings in the embodiments of the present invention.

[0034] Example 1:

[0035] like Figure 1 As shown, this embodiment provides an optical feature-driven method for identifying printed matter, including the following steps:

[0036] S101, Obtain the original color set generated by uniform sampling within the specified color gamut;

[0037] S102, the original color set is input into the pre-trained color set prediction model under various printing product classification types to obtain the predicted printed color set under the corresponding printing product classification type. The color set prediction model is pre-trained to establish the mapping relationship between the input original color set and the output predicted printed color set under the printing product classification type.

[0038] S103, using the similarity between the color set of the identified printed matter and the predicted post-printing color set under each printed matter classification type to determine the printed matter classification type to which the identified printed matter corresponds.

[0039] In step S101 of the embodiment, when the original color set generated by uniform sampling in the specified color domain is obtained, the specified color domain can be a desired color domain, such as an RGB color domain, a CMYK color domain, and an HSI color domain, according to actual needs. As an optional implementation, the specified color domain in the embodiment refers to an RGB color domain, and obtaining the original color set generated by uniform sampling in the specified color domain includes performing equal-interval sampling on N value points in the red channel R, the green channel G, and the blue channel B of the RGB color domain, respectively, to obtain an original color set composed of N x N x N color value points. For example, as an optional implementation, 25 value points are sampled in the R, G, and B channels of the RGB color domain, respectively, to generate 25 x 25 x 25 = 15625 colors, forming a digital color card covering the entire RGB color domain, and the digital colors constitute the original color set A.

[0040] The plurality of printed matter classification types in step S102 of the embodiment include original printed matter and copied printed matter, where the original printed matter refers to printed matter obtained by printing or printing for the first time according to a design drawing, and the copied printed matter refers to printed matter obtained by printing or printing after scanning based on the original printed matter or a copy of the original printed matter.

[0041] To solve the problem that it is difficult for traditional manual identification to quantify the subtle color difference in the printing and scanning and reprinting process, the embodiment further includes training of a color set prediction model under various printed matter classification types before step S102, and the training step of the color set prediction model under any printed matter classification type includes:

[0042] S201, obtaining a printed matter under the current printed matter classification type of the digital color card of the original color set A, which is obtained by printing or printing of the digital color card of the original color set under the current printed matter classification type; in the embodiment, specifically, the digital color card of 15625 colors is obtained by printing as a paper color card of the original printed matter, and then the paper color card of the original printed matter is scanned and printed to generate a paper color card of the copied printed matter;

[0043] S202, using a colorimeter to detect the printed color set of the printed matter, as an optional embodiment, ten scans are performed in each color area in the embodiment to take the average value to eliminate the interference of local extreme value and random measurement error; in the embodiment, the measured printed color set A1 of the original printed matter paper color card; the measured printed color set A2 of the paper color card of the copied printed matter;

[0044] S203, normalize the printed color set as a printed color set sample, by normalizing the RGB values recorded in the printed color set to the range of [0, 1], the influence of the difference between different channel value ranges on the subsequent color set prediction model training can be eliminated;

[0045] S204, construct a training data set of the original color set and the color set sample of the corresponding printed matter, for example, in the embodiment, the data is divided into a training set, a validation set and a test set according to the ratio of 7:2:1, to provide a standardized data basis for subsequent model construction and evaluation;

[0046] S205, establish a color set prediction model under the classification type of the current printed matter, the input of the color set prediction model is the original color set, and the output is the predicted printed color set, the color set prediction model is trained by using the training data set, so as to establish the mapping relationship between the input original color set and the output predicted printed color set under the classification type of the printed matter, thereby effectively quantifying the subtle color difference between the original printed matter and the copied printed matter.

[0047] The color set prediction model in step S102 of the embodiment can use the required deep learning model as needed, for example, as an optional embodiment, a support vector regression (SVR) model is used in the embodiment. The support vector regression model is used as the basic model, and the single-output support vector regression model is expanded to three-channel output by a multi-output regressor, which corresponds to the RGB three channels of the predicted color respectively. In the model training stage, as shown in Figure 2 two mapping models A→A1 and A→A2 are constructed respectively, the grid search is performed on the key parameters (such as penalty coefficient C, insensitive loss parameter epsilon, kernel function parameter gamma, etc.) through five-fold cross-validation, the optimal parameter combination that minimizes the mean square error (MSE) is screened out, the smaller the mean square error is, the better the simulation effect of the model on the printing color difference is, and finally the optimal color set prediction model SVR1 (corresponding to A→A1 mapping) and the color set prediction model SVR2 (corresponding to A→A2 mapping) are obtained, the two models can respectively simulate the color deviation law in the single printing of the original printed matter and the secondary printing (scanning and printing again) of the copied printed matter with high precision, and provide a reliable prediction benchmark for the subsequent type identification of the original printed matter and the copied printed matter. Finally, asFigure 3 As shown, the original color set A is input into the color set prediction model SVR1 corresponding to the original printed matter to obtain the predicted post-printing color set A1' under the classification type of the corresponding printed matter, and the original color set A is input into the color set prediction model SVR2 corresponding to the copy printed matter to obtain the predicted post-printing color set A2' under the classification type of the corresponding printed matter, which can be used for the classification of the identified printed matter.

[0048] In step S103, the similarity of the color set of the identified printed matter and the predicted post-printing color set under each printed matter classification type is used to determine the printed matter classification type corresponding to the identified printed matter. According to the needs, a similarity function, a geometric distance function, a correlation function, and a deep learning model can be used to determine the printed matter classification type corresponding to the identified printed matter by using the similarity of the color set of the identified printed matter and the predicted post-printing color set under each printed matter classification type. As an optional embodiment, as shown in Figure 4 As shown, the similarity of the color set of the identified printed matter and the predicted post-printing color set under each printed matter classification type is used to determine the printed matter classification type corresponding to the identified printed matter in step S103 of the embodiment, which includes:

[0049] S301, obtaining the color set of the identified printed matter;

[0050] S302, inputting the color set of the identified printed matter and the predicted post-printing color set under each printed matter classification type (including the predicted post-printing color set A1' and the predicted post-printing color set A2' in the embodiment) as input data into the pre-trained printed matter identification model, obtaining the similarity scores of the color set of the identified printed matter and the predicted post-printing color set under each printed matter classification type (similarity score 1 and similarity score 2 in the embodiment) through the printed matter identification model, and taking the printed matter classification type with the largest similarity score as the determined printed matter classification type corresponding to the identified printed matter.

[0051] To solve the problem that the traditional method relies on manual experience to judge similarity and is difficult to automatically distinguish between the first printing original and the scanned copy, the printed matter identification model in step S103 in the embodiment is a twin network, which maps the two color values of the photographed color sample and the color set predicted by the color set prediction model to the same feature space through a shared feature extractor, calculates and compares the feature similarity, thereby automatically judging the printing type of the target pattern, and realizing intelligent identification without manual intervention. Specifically, the twin network in the embodiment includes a shared feature extractor and a plurality of independent classification heads, the feature extractor is used to extract the color set of the identified printed matter or the printed color set under a certain printed matter classification type to obtain an embedding vector representing the color set feature, and the number of classification heads is the same as the number of printed matter classification types. Each classification head is used to generate a similarity score between the embedding vectors of the color set of the identified printed matter and the printed color set under the corresponding printed matter classification type. Based on the prediction result of the color set prediction model, a sample pair (same class / different class) of the twin network can be constructed, and the similarity discrimination rule in the feature space is learned through the cooperative training of the contrast loss and the binary cross entropy loss. Finally, the printing type of the target pattern is output according to the feature similarity.

[0052] As an optional implementation, the feature extractor in the twin network in the embodiment includes a three-layer fully connected network and a splicing module, the three-layer fully connected network is used to extract the color set of the identified printed matter or the printed color set under a certain printed matter classification type to obtain a 16-dimensional embedding vector representing the color set feature, and the splicing module is used to splice the embedding vectors of the color set of the identified printed matter and the printed color set under the corresponding printed matter classification type to obtain a spliced embedding vector; the classification head in the twin network is composed of a two-layer fully connected network, which is used to map the spliced embedding vector to a similarity score; and the total loss function adopted by the twin network during training is the weighted sum of the contrast loss and the binary cross entropy loss. In the classification effect evaluation, the classification accuracy of the twin network on the test set is mainly used to comprehensively verify the identification performance of the model. Among them, the accuracy reflects the overall classification ability of the model, the average distance and the distance ratio reflect the ability of the model to distinguish between same class and different class sample pairs, and the greater the distance ratio indicates that the discrimination boundary of the model is clearer and the identification robustness is stronger. Through the above multi-dimensional evaluation indexes, the effectiveness and reliability of the method of the present application in the printing color difference simulation and the copy identification task are comprehensively verified.

[0053] To verify the technical effect of the optical feature driven printed matter identification method of the embodiment, the experiment is carried out based on a test set containing 1562 samples, various color features of the RGB color domain are fully covered, and the representativeness and reliability of the verification result are ensured. In the color prediction task, the MSE of the color set prediction model SVR1 is 0.001183, and the MSE of the color set prediction model SVR2 is 0.001789, both of which indicate that the support vector regression model adopted by the color set prediction model has high prediction accuracy on the test set and has very high generalization ability. In the classification identification task, the experiment compares the feature distance-based classifier with the twin network-based classifier, and the final result is that the accuracy of the distance-based classifier is 68.37%, while the accuracy of the twin network-based classifier of the optical feature driven printed matter identification method of the embodiment is 92.32%. The above experimental results fully prove the rationality and effectiveness of the optical feature driven printed matter identification method of the embodiment, and the optical feature driven printed matter identification method of the embodiment has strong feature learning and class division ability, can accurately capture the key differences between samples, provides strong support for distinguishing printing and scanning reprinting, has important practical significance for quickly identifying scanning-printing counterfeit behavior and curbing illegal counterfeiting.

[0054] In addition, the embodiment also provides an optical feature driven printed matter identification system, which comprises a microprocessor and a memory connected with each other, and the microprocessor is programmed or configured to execute the optical feature driven printed matter identification method. The embodiment also provides a computer readable storage medium, which stores a computer program or instructions programmed or configured to execute the optical feature driven printed matter identification method by a processor. The embodiment also provides a computer program product comprising a computer program or instructions programmed or configured to execute the optical feature driven printed matter identification method by a processor.

[0055] Embodiment two:

[0056] The embodiment is basically the same as embodiment one, and the main difference is that the classification types of the plurality of printed matters in step S102 are different. The classification types of the plurality of printed matters in step S102 of the embodiment include qualified original printed matters and unqualified original printed matters. The unqualified original printed matters are original printed matters whose quality is unqualified due to factors such as unqualified printing equipment and unqualified printing process. The original printed matters refer to printed matters obtained by printing or printing according to a design drawing for the first time. In step S201, the digital color card of the original color set A is obtained under the current classification type of the printed matters. The printed matters are obtained by printing or printing according to the digital color card of the original color set under the current classification type of the printed matters. In the embodiment, the digital color card of 15625 colors is obtained by printing on paper color cards as qualified original printed matters and unqualified original printed matters by qualified and unqualified printing equipment / printing process, respectively. The remaining methods are basically the same as those in embodiment one, and thus will not be described herein. The optical feature driven printed matter identification method of the embodiment has strong feature learning and category division capability, can accurately capture the key differences between samples, and provides strong support for distinguishing qualified original printed matters and unqualified original printed matters. The optical feature driven printed matter identification method has important practical significance for quickly identifying qualified original printed matters and unqualified original printed matters and ensuring printing quality.

[0057] In addition, the embodiment also provides an optical feature driven printed matter identification system, which includes a microprocessor and a memory connected to each other. The microprocessor is programmed or configured to execute the optical feature driven printed matter identification method. The embodiment also provides a computer readable storage medium, which stores a computer program or instructions programmed or configured to execute the optical feature driven printed matter identification method by a processor. The embodiment also provides a computer program product, which includes a computer program or instructions programmed or configured to execute the optical feature driven printed matter identification method by a processor.

[0058] Those skilled in the art will appreciate that the technology of the present application can be embodied in a method, system, or computer program product. Accordingly, the present application can take the form of an entirely hardware embodiment, an entirely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present application can take the form of a computer program product on one or more computer-readable storage media (including, but not limited to, disk memory, CD-ROMs, optical storage devices, etc.) embodying computer usable program code. The specification and drawings are, accordingly, to be regarded as illustrative and not restrictive. The present application is described in terms of flowcharts and / or block diagrams in which each block indicates a set of elements carried out normally by a computer program. Each block and combinations thereof can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general purpose computer, special purpose computer, embedded processing system, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, create means for implementing the functions specified in the flowchart or block diagram block or blocks. These computer program instructions can also be stored in a computer readable memory that can direct a computer or other programmable data processing apparatus to function in a particular manner, such that the instructions stored in the computer readable memory produce an article of manufacture including instructions which implement the function specified in the flowchart or block diagram block or blocks. The computer program instructions can also be loaded onto a computer or other programmable data processing apparatus to cause a series of operational steps to be performed on the computer or other programmable apparatus to produce a computer implemented process such that the instructions which execute on the computer or other programmable apparatus provide steps for implementing the functions specified in the flowchart or block diagram block or blocks. Alternatively, computer program Figure 1 Figure 1 Figure 1 Figure 1 Figure 1 Figure 1

[0059] The above description is only the preferred embodiment of the present application, the protection scope of the present application is not limited to the above-mentioned embodiment, any technical solution falling into the idea of the present application is within the protection scope of the present application. It should be noted that, for the ordinary skilled in the art, some improvements and refinements without departing from the principle of the present application are also considered as the protection scope of the present application.​​​​​​

Claims

1. An optical feature driven print authentication method, characterized by, The method comprises the following steps: S101, obtaining an original color set generated by uniform sampling in a specified color domain; S102, inputting the original color set into a color set prediction model pre-trained under a plurality of printing product classification types to obtain a predicted printed color set under a corresponding printing product classification type, wherein the color set prediction model is pre-trained to establish a mapping relationship between an input original color set and a predicted printed color set under the printing product classification type; S103, taking the color set as an optical feature of the printing product, and inputting the optical feature into a pre-trained printing product identification model to determine the printing product classification type corresponding to the identified printing product by using the similarity between the color set of the identified printing product and the predicted printed color set under each printing product classification type; The printing product identification model is a twin network; The plurality of printing product classification types in step S102 include original printing products and copied printing products, or the plurality of printing product classification types in step S102 include qualified original printing products and unqualified original printing products; The original printing product refers to a printing product obtained by printing or printing for the first time according to a design drawing, and the copied printing product refers to a printing product obtained by printing or printing after scanning based on an original printing product or a copy of the original printing product.

2. The method of claim 1, wherein the optical feature is a fluorescent feature. In step S101, the specified color domain refers to an RGB color domain, and the original color set generated by uniform sampling in the specified color domain comprises equally spaced sampling N value points on the red channel R, the green channel G and the blue channel B of the RGB color domain, thereby obtaining an original color set composed of N×N×N color different value points.

3. The method of claim 1, wherein the optical feature is a fluorescent feature. Before step S102, the color set prediction model under each printing product classification type is trained, and the training steps of the color set prediction model under any printing product classification type comprise: S201, obtaining a printing product of a digital color card of the original color set under the current printing product classification type, wherein the printing product is obtained by printing or printing of the digital color card of the original color set under the current printing product classification type; S202, detecting and obtaining the printed color set of the printing product by using a colorimeter; S203, normalizing the printed color set as a printed color set sample; S204, constructing a training data set of the original color set and the color set sample of the corresponding printing product; S205, establishing a color set prediction model under the current printing product classification type, wherein the input of the color set prediction model is the original color set, and the output is the predicted printed color set, and the color set prediction model is trained by using the training data set, so as to establish a mapping relationship between the input original color set and the predicted printed color set under the printing product classification type.

4. The method of claim 1, wherein the optical feature is a fluorescent feature. In step S103, the similarity between the color set of the identified printing product and the predicted printed color set under each printing product classification type is used to determine the printing product classification type corresponding to the identified printing product, which comprises: S301, obtaining the color set of the identified printing product; S302, input the color set of the to-be-authenticated printed matter and the predicted post-printing color set under each printed matter classification type as input data into the pre-trained printed matter authentication model, obtain the similarity score of the color set of the to-be-authenticated printed matter and the predicted post-printing color set under each printed matter classification type through the printed matter authentication model, and determine the printed matter classification type with the largest similarity score as the printed matter classification type corresponding to the to-be-authenticated printed matter.

5. The method of claim 4, wherein the optical feature is a fluorescent feature. The twin network comprises a shared feature extractor and a plurality of independent classification heads, the feature extractor is configured to extract the color set of the to-be-authenticated printed matter or the post-printing color set under a certain printed matter classification type to obtain an embedding vector representing color set features, and the number of the classification heads is the same as the number of printed matter classification types, each classification head is configured to generate a similarity score between the embedding vectors of the color set of the to-be-authenticated printed matter and the post-printing color set under the corresponding printed matter classification type.

6. The method of claim 5, wherein the optical feature is a fluorescent feature. The feature extractor in the twin network comprises a three-layer fully connected network and a splicing module, the three-layer fully connected network is configured to extract the color set of the to-be-authenticated printed matter or the post-printing color set under a certain printed matter classification type to obtain a 16-dimensional embedding vector representing color set features, and the splicing module is configured to splice the embedding vectors of the color set of the to-be-authenticated printed matter and the post-printing color set under the corresponding printed matter classification type to obtain a spliced embedding vector; the classification head in the twin network is composed of a two-layer fully connected network, configured to map the spliced embedding vector to a similarity score; and the total loss function adopted by the twin network during training is the weighted sum of the contrastive loss and the binary cross-entropy loss.

7. An optical feature driven printed matter authentication system comprising a microprocessor and a memory connected to each other, characterized in that, The microprocessor is programmed or configured to perform the printed matter authentication method driven by the optical features according to any one of claims 1-6.

8. A computer-readable storage medium having stored therein a computer program or instructions, characterized in that, The computer program or instructions are programmed or configured to perform the printed matter authentication method driven by the optical features according to any one of claims 1-6 through the processor.

9. A computer program product comprising computer programs or instructions, characterized in that, The computer program or instructions are programmed or configured to perform the printed matter authentication method driven by the optical features according to any one of claims 1-6 through the processor.

Images