Certificate identification model training method, certificate identification method, equipment, medium and product

Abstract

The invention provides a certificate identification model training method and device, a certificate identification method and device, a medium and a product. The certificate identification model training method comprises the steps that a certificate sample set mixed with multiple certificate types is acquired, certificate authenticity labels are marked on multi-modal certificate samples included in the certificate sample set, and the multi-modal certificate samples comprise certificate images and certificate anti-counterfeiting rules; the certificate anti-counterfeiting rule of any certificate type is used for describing the position of at least one anti-counterfeiting element of the certificate type in the certificate and the anti-counterfeiting rule of the anti-counterfeiting element; inputting the multi-modal certificate sample into a to-be-trained certificate identification model, performing identification by the to-be-trained certificate identification model based on the certificate image and the certificate anti-counterfeiting rule, and outputting a certificate authenticity judgment result; and training the certificate identification model by taking minimization of an error between a certificate authenticity label and a certificate authenticity judgment result as an optimization target. And high precision, high generalization capability and expandability of certificate authenticity identification are realized.

Description

TECHNICAL FIELD

[0001] One or more embodiments of the present specification relate to the technical field of artificial intelligence, and in particular, to a certificate identification model training method, a certificate identification method, an electronic device, a computer readable storage medium, and a computer program product. BACKGROUND

[0002] With the popularity of electronic identity authentication (eKYC, electronic Know Your Customer), financial, Internet, entry and exit management, and other service scenarios generally need to provide remote verification capabilities for global customers. In actual applications, various types of certificates worldwide need to be identified for authenticity. However, there are a large number of certificate types worldwide, and certificates in different countries and regions differ significantly. According to statistics, there are tens of thousands of common certificate types worldwide.

[0003] Traditional certificate anti-counterfeiting detection usually relies on a classification model based on machine learning, which needs to train a corresponding high-imitation detection model for each certificate type to complete authenticity identification. However, this method is difficult to cover all certificate types in actual scenarios. Since the model is trained for a specific certificate type, when facing a new certificate type that has not been seen before, it is often difficult to accurately complete authenticity judgment, and it is difficult to meet the needs of global service scenarios. SUMMARY

[0004] Therefore, one or more embodiments of the present specification provide technical solutions as follows: According to a first aspect of one or more embodiments of the present specification, a certificate identification model training method is provided, including: Obtaining a certificate sample set mixed with multiple certificate types, the certificate sample set including multi-modal certificate samples labeled with certificate authenticity labels, the multi-modal certificate samples including certificate images and certificate anti-counterfeiting rules, and the certificate anti-counterfeiting rules of any certificate type being used to describe the position of at least one anti-counterfeiting element of the certificate type in the certificate and the anti-counterfeiting rules of the anti-counterfeiting element; Inputting the multi-modal certificate samples into a certificate identification model to be trained, so that the certificate identification model to be trained performs identification based on the certificate images and the certificate anti-counterfeiting rules, and outputs a certificate authenticity judgment result; Training the certificate identification model with the optimization target of minimizing the error between the certificate authenticity labels and the certificate authenticity judgment result.

[0005] According to a second aspect of one or more embodiments of the present specification, a certificate identification method is provided, including: Obtaining a certificate image to be identified, and determining the certificate type to which the certificate image to be identified belongs; query a matched certificate anti-counterfeiting rule from a certificate anti-counterfeiting rule library based on the to-be-identified certificate image and / or the certificate type thereof; input the to-be-identified certificate image and the certificate anti-counterfeiting rule into a trained certificate identification model, and obtain a certificate authenticity judgment result output by the certificate identification model; wherein the certificate identification model is trained based on the method of the first aspect.

[0006] According to a third aspect of an embodiment of the present specification, an electronic device is provided, comprising: a processor; a memory for storing processor-executable instructions; wherein the processor, when executing the executable instructions, is configured to implement the method of the first aspect.

[0007] According to a fourth aspect of an embodiment of the present specification, a computer-readable storage medium is provided, which stores a computer program, and the program, when executed by a processor, implements the steps of the method of the first aspect.

[0008] According to a fifth aspect of an embodiment of the present specification, a computer program product is provided, comprising a computer program, and the computer program, when executed by a processor, implements the steps of the method of the first aspect.

[0009] As can be seen from the above embodiments, the present specification trains a model by using multi-modal certificate samples (containing certificate images and corresponding types of anti-counterfeiting rules) of mixed multiple certificate types, and outputs a true or false judgment result based on the association between the certificate image and the anti-counterfeiting rule during the model training process, so that the model can learn the matching logic between the image features of different certificate types and the anti-counterfeiting rules, break through the limitations of single certificate type training, improve the adaptation ability to multiple certificate types, and especially enhance the generalization recognition ability to new certificate types that have not been seen before; with the fusion analysis of multi-modal information (image and anti-counterfeiting rule), the model can more accurately capture the core judgment basis of certificate authenticity, and improve the accuracy of true or false identification; without training a model for each certificate, the mixed training of a unified model reduces the model training cost and management complexity in the global certificate identification scene.

[0010] It should be understood that the above general description and the following detailed description are only exemplary and explanatory, and cannot limit the present specification. BRIEF DESCRIPTION OF DRAWINGS

[0011] Figure 1 is a flowchart of a certificate identification model training method provided by an exemplary embodiment.

[0012] Figure 2 is a schematic diagram of a certificate image and its corresponding certificate anti-counterfeiting rule provided by an exemplary embodiment.

[0013] Figure 3 is a schematic diagram of an original certificate image and a processed certificate image provided by an example embodiment.

[0014] Figure 4 is a flowchart of a certificate identification method provided by an example embodiment.

[0015] Figure 5 is a schematic diagram of certificate identification model input and output provided by an example embodiment.

[0016] Figure 6 is a structural schematic diagram of a device provided by an example embodiment. DETAILED DESCRIPTION

[0017] In order for those skilled in the art to better understand the technical solutions in the specification, the technical solutions in the specification will be clearly and completely described below in conjunction with the drawings in the specification. Obviously, the described embodiments are only some of the embodiments of the specification, not all the embodiments. Based on the embodiments in the specification, all other embodiments obtained by those of ordinary skill in the art without creative labor should be within the scope of protection of the specification.

[0018] The user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data for analysis, stored data, displayed data, etc.) involved in the specification are information and data authorized by the user or authorized by all parties, and the collection, use and processing of related data need to comply with relevant laws, regulations and standards of relevant countries and regions, and provide corresponding operation portal for user to choose authorization or refusal.

[0019] Electronic identity authentication (eKYC, electronic Know Your Customer) provides customer verification capability for global customers, which needs to identify global certificates, and there are tens of thousands of global certificates, and the anti-fake marks and layouts of each certificate are different. It is not efficient to train a high-imitation detection model for each certificate, and it is also difficult to implement.

[0020] Based on this, please refer to Figure 1 The embodiment of the specification provides a certificate identification model training method, which can be executed by an electronic device, including but not limited to a server (such as a physical server, a virtual server, etc.), a smart phone / mobile phone, a tablet computer, a personal digital assistant (PDA), a laptop computer, a desktop computer, a wearable device or other types of devices. The training method includes: In S100, a set of certificate samples of multiple types of certificates is obtained, the multi-modal certificate samples included in the set of certificate samples are labeled with certificate authenticity labels, and the multi-modal certificate samples include certificate images and certificate anti-counterfeiting rules. Any certificate type certificate anti-counterfeiting rule is used to describe the position of at least one anti-counterfeiting element of the certificate type in the certificate and the anti-counterfeiting rule of the anti-counterfeiting element.

[0021] Among them, the certificate anti-counterfeiting rule is used to describe the anti-counterfeiting characteristics inherent in a certain type of certificate. Its content not only includes the anti-counterfeiting element itself, but also includes the position, form of expression, structural characteristics and coding rules of the element in the certificate. By defining and extracting these anti-counterfeiting rules in advance, text or structured auxiliary information can be provided for the certificate identification model, so that the model can compare image features with rule features in cross-modal comparison when judging the authenticity of the certificate, thereby improving the identification accuracy. Please refer to Figure 2 , which shows a certificate image and its corresponding certificate anti-counterfeiting rule.

[0022] For example, the certificate anti-counterfeiting rule can include but is not limited to: ① Watermark rule: describes the distribution position of the watermark in the certificate, such as the middle of the front or the side area; describes the presentation mode of the watermark, such as the alternating shadow, the translucent visible, the relief effect; specifies the shape or content of the watermark, such as the institution logo, the specific pattern.

[0023] ② Safety line rule: describes the existence position of the embedded safety line in the certificate, such as horizontally or vertically through a certain area; indicates the type of safety line, such as full-embedded, exposed, and windowed; defines the visible features of the safety line, such as magnetism, fluorescence reaction, and color change.

[0024] ③ Anti-counterfeiting pattern rule: describes the background pattern, geometric texture or hologram of a specific area of the certificate; specifies the complexity of the anti-counterfeiting pattern, such as micro-text, gradient line, and optical variable image; determines the performance characteristics under magnification or specific light source.

[0025] ④ Font and layout rule: defines the special font used in the certificate, such as monospaced font and special symbol font; describes the arrangement rule of the number or text, such as the length of the certificate number, the segmentation method, and the combination mode of letters and numbers; includes the anti-counterfeiting characteristics of the font, such as laser engraving, relief printing, and micro-text.

[0026] ⑤ Paper and material rule: describes the physical characteristics of the paper or substrate used in the certificate, such as fiber distribution, touch, and fluorescent fiber; includes the safety components embedded in the material, such as magnetic particles and chemical anti-counterfeiting agents.

[0027] ⑥ Texture and area rule: describes the texture characteristics of a specific area, such as the complexity of the texture of the portrait area; specifies the distribution rule of the micro-character or micro-symbol in the local area.

[0028] (7) Encoding and sequence rules: define the generation rules of the certificate number, such as region code + birth month + sequence code + check digit; describe the coding rules or date format of the valid period of the issuing agency; explain whether it has a specific verification mechanism, such as the check digit calculation formula.

[0029] By converting the anti-counterfeiting features of various certificates into unified rule descriptions, the model can reuse the training framework across different certificate types, avoid relying solely on image learning, and reduce training complexity. Certificate anti-counterfeiting rules as text or structured input are input into the model together with certificate images, so that the model can "look at the image" and "check the rules" when determining authenticity, achieving cross-modal alignment.

[0030] Among them, the certificate image can include a real certificate image and a counterfeit certificate image, so as to improve the model's ability to distinguish between true and false during training. By constructing a sample set that covers multiple certificate types and contains both images and certificate anti-counterfeiting rules, the diversity and representativeness of the training data can be ensured, thereby supporting the subsequent model to have cross-type generalization ability.

[0031] In S102, the multi-modal certificate sample is input into the certificate identification model to be trained, so that the certificate identification model to be trained identifies based on the certificate image and the certificate anti-counterfeiting rule, and outputs a certificate authenticity judgment result.

[0032] Illustratively, the certificate identification model to be trained is a multi-modal large model that can simultaneously process image features and text features and jointly model based on the two types of features. For example, the certificate identification model can include an image encoding subnetwork and a rule encoding subnetwork, the former for extracting local texture features and overall layout features of the certificate image to obtain image features, and the latter for encoding the text or structured description of the certificate anti-counterfeiting rule into a vector representation to obtain rule features. The model can also include a feature fusion module for mapping image features and rule features to a unified semantic space, thereby achieving cross-modal alignment and joint reasoning, and outputting a certificate authenticity judgment result.

[0033] By inputting the certificate image and the certificate anti-counterfeiting rule into the certificate identification model, the reasoning process of humans when distinguishing the authenticity of the certificate can be simulated, that is, observing the image content and relying on the anti-counterfeiting rule. This not only improves the accuracy of identification, but also enables the model to have better generalization ability when facing new certificate types.

[0034] In S104, the certificate identification model is trained with the optimization goal of minimizing the error between the certificate authenticity label and the certificate authenticity judgment result.

[0035] By way of example, the error between the certificate authenticity label and the certificate authenticity judgment result can be quantified by a cross-entropy loss function, a mean square error function, or other classification loss functions. In each training iteration, the certificate identification model first obtains a prediction result (i.e., a certificate authenticity judgment result) through a forward propagation process, and then calculates the gradient of the error with respect to the model parameters based on the value of the loss function. The gradient can be efficiently calculated by a backpropagation algorithm, which ensures that the error information is passed back to the model parameters layer by layer. In the parameter updating process, optimization algorithms such as gradient descent, stochastic gradient descent, Adam, etc. can be used to gradually adjust the weight parameters in the model. The goal of the updating process is to gradually reduce the value of the loss function, so that the output result of the model is closer to the true label. The training process is iteratively performed until a preset convergence condition is reached, for example, the training error is less than a certain threshold, or the performance on the validation set reaches a stable state. During the iteration process, batch training can be used to improve the training efficiency and the generalization ability of the model. Through the above training process based on error minimization, the model can gradually learn the corresponding relationship between the certificate image and the anti-fake rule, and strengthen its discriminant ability in multi-modal feature fusion.

[0036] In an implementation, the trained certificate identification model can be deployed in a cloud server to process large-scale certificate sample data with the help of a distributed training framework, or it can be deployed on an edge device for lightweight inference to support real-time certificate identification on mobile devices.

[0037] This embodiment trains by obtaining a certificate sample set that mixes multiple certificate types, and introduces multi-modal information of certificate images and certificate anti-fake rules. The certificate identification model can learn the commonalities and differences between different certificate types in images and anti-fake rules, which can improve the accuracy of certificate identification and also has certain authenticity judgment ability for newly emerging certificate types that are not explicitly present in the training set, effectively improving the generalization performance of the model and greatly expanding its application scope.

[0038] In some embodiments, the identification process can also be assisted by a fake certificate image library to further improve the discriminant ability of the certificate identification model in high-imitation fake certificate scenarios. By way of example, the certificate images in the multi-modal certificate sample can include: a real certificate image, and a fake certificate image with a similarity greater than a preset threshold to the real certificate image obtained from the fake certificate image library. By introducing high-similarity real and fake certificate images, the model can directly learn the subtle differences between the two types of images, thereby more accurately corresponding the image details to the anti-fake rules and breaking through the recognition blind area in high-imitation scenarios.

[0039] In some embodiments, the multi-modal certificate sample includes positive samples and negative samples to enhance the discriminant ability of the model during the training process.

[0040] The certificate image and the certificate anti-counterfeiting rule in the positive sample have a matching relationship. For example, the positive sample includes a certificate image belonging to any certificate type and a correct anti-counterfeiting rule corresponding to the certificate type; for example, a picture of an ID card + a correct ID card coding rule, a watermark position description, etc. The role of the positive sample is to provide a real and valid feature correspondence relationship to the model, so that the model can learn the one-to-one correspondence between the image features and the rule description under different certificate types.

[0041] The certificate image and the certificate anti-counterfeiting rule in the negative sample do not have a matching relationship. For example, the negative sample includes at least one of the following: ① a certificate image belonging to any certificate type and an incorrect anti-counterfeiting rule corresponding to the certificate type; for example, a picture of an ID card + an incorrect ID card coding rule, an incorrect watermark position description, etc. ② a certificate image belonging to a second certificate type and a correct anti-counterfeiting rule corresponding to a first certificate type; for example, a picture of a passport + an ID card coding rule. ③ a certificate image belonging to a first certificate type and a correct anti-counterfeiting rule corresponding to a second certificate type; for example, a picture of an ID card + a passport coding rule. The first certificate type and the second certificate type belong to different types. Through the design of the above negative sample, different types of mismatching situations can be fully covered, so that the model can learn how to distinguish between real corresponding relationships and fake corresponding relationships during training.

[0042] During the training of the above certificate identification model, the model not only minimizes the error between the certificate authenticity label and the prediction result as the optimization target, but also further introduces a contrastive loss constraint. For example, for the positive sample, the optimization target includes minimizing the distance between the features extracted by the certificate identification model from the certificate image and the certificate anti-counterfeiting rule in the positive sample, i.e., as close as possible to the semantic representation of the two features, to ensure that the image information and the rule information are consistent in the feature space. For the negative sample, the optimization target includes maximizing the distance between the features extracted by the certificate identification model from the certificate image and the certificate anti-counterfeiting rule in the negative sample, i.e., effectively distinguishing the mismatched image features and rule features in the feature space. These two optimization targets can be achieved through a contrastive loss function, a triplet loss function, or other metric learning loss functions, thereby improving the discriminative ability of the model in multi-modal feature fusion.

[0043] By introducing positive and negative sample pairs for training, the model can not only learn the appearance features of the certificate image itself, but also capture the logical correspondence between the image features and the anti-fake rules. This mechanism can effectively avoid the model relying only on visual representation for memory, thereby improving its recognition ability of fine-grained features. For example, the model can distinguish between "the ID card rule is correct but the image belongs to a passport" or "the image is real but the rule description is wrong". Finally, the trained model has stronger robustness and generalization ability, and can accurately identify the authenticity of different certificates in different countries and regions.

[0044] It should be noted that the division of positive and negative samples is not only based on whether the certificate image is a real image, but also based on whether the certificate image and the anti-fake rule have a matching relationship.

[0045] The positive sample can include a combination of a real certificate image and the correct anti-fake rule of the certificate type. For example, a real ID card image matches the real rule of ID card encoding rule, watermark position description, etc.

[0046] In addition, the positive sample can also include a combination of a fake certificate image and the corresponding fake rule. For example, in some high-imitation fake certificates, the forger may construct anti-fake description information consistent with the fake image, at which time the fake certificate image and the corresponding fake rule still have a matching relationship and belong to the positive sample.

[0047] The negative sample is a sample that does not have a matching relationship between the certificate image and the anti-fake rule. For example, a real ID card image combined with the number rule of a passport, or a fake passport image combined with the number rule of an ID card. Such samples are used in training to prompt the model to learn to distinguish mismatched features, thereby improving the discrimination ability.

[0048] Through this design, the model can not only utilize the diversity of real certificate images and fake certificate images during training, but also learn the core judgment logic of "whether the image features and rule features are consistent". This scheme can effectively avoid the model's over-reliance on image appearance features, and instead guide the model to establish a cross-modal correspondence, thereby improving its generalization ability and robustness in a global range of multiple certificate types.

[0049] In a possible implementation, the correct anti-counterfeiting rule can be obtained in the following manner: when a certain certificate image is acquired, the type of the certificate to which the certificate image belongs is first determined, and then a query is performed in a pre-constructed certificate anti-counterfeiting rule library based on the certificate type, so as to obtain the correct anti-counterfeiting rule corresponding to the certificate type. The certificate anti-counterfeiting rule library can prestore anti-counterfeiting rule information related to multiple certificate types, and the types of the certificate types involved are more than the types of the certificate types participating in the model training. In other words, the certificate types used in the training process are only a part of the anti-counterfeiting rule library, and the anti-counterfeiting rule library can cover a larger range of certificate types.

[0050] In the embodiments of the present application, the model can realize zero-shot learning in this way. Zero-shot learning means that the model has not been exposed to a certain target type of certificate sample during the training process, but can still predict the authenticity of the target type of certificate by using the matching relationship between the certificate image features and the anti-counterfeiting rule features in the reasoning stage. This embodiment does not need to collect additional samples and retrain the model for new certificate types, and only needs to configure the rule information of the certificate type in the certificate anti-counterfeiting rule library, and the model can realize the authenticity identification of the certificate type.

[0051] In the above manner, the present embodiment can significantly improve the expansibility and generalization ability of the model, and reduce the cost of maintaining and updating the certificate identification system in the global range. When a new certificate type is encountered, only the rule description of the certificate type needs to be supplemented in the certificate anti-counterfeiting rule library, so that the model has the ability to identify the authenticity of the certificate, without the need for large-scale sample collection and model training, thereby meeting the demand for rapid adaptation of diversified certificates in eKYC and other cross-border identity verification scenarios.

[0052] In another possible implementation, instead of using zero-shot learning, a small amount of real samples and fake samples can be collected for each new certificate type, and the certificate identification model is fine-tuned using a small amount of samples to realize rapid adaptation based on few-shot. In this embodiment, the model can obtain higher identification accuracy for a specific certificate type through Few-Shot learning, because the model further updates the parameters based on the specific features of the certificate type on the basis of the original general identification ability. This way maintains the training efficiency while improving the identification performance of some new certificate types.

[0053] In some embodiments, the multi-modal certificate sample set includes multi-modal certificate samples that are also labeled with the basis for the determination corresponding to the certificate authenticity label. For example, when an expected anti-counterfeiting element is missing or there is an encoding error in a counterfeit certificate image, the "missing anti-counterfeiting element, encoding error" can be labeled as the basis for determining that the certificate is fake; when the watermark position of a real certificate meets the pre-set rules, and the identity card meets the coding specification, the "watermark position meets the rules, number coding meets the rules" can be labeled as the basis for determining that the certificate is real; for example, when the watermark position of a certificate does not meet the rules, the "watermark position does not meet the rules" can be labeled as the basis for determining that the certificate is fake.

[0054] In this embodiment, the certificate identification model outputs the basis for the determination corresponding to the certificate authenticity determination result when outputting the certificate authenticity determination result. Thus, the optimization target of the model includes not only minimizing the error between the certificate authenticity label and the certificate authenticity determination result, but also minimizing the error between the basis for the determination corresponding to the certificate authenticity label and the basis for the determination corresponding to the certificate authenticity determination result. Through such a training method, the model can not only maintain the accuracy of authenticity identification, but also output the basis for the determination with interpretability, thereby improving the usability and credibility of the model in actual application.

[0055] In other embodiments, the certificate samples in the multi-modal certificate sample set can not be labeled with the basis for the determination corresponding to the certificate authenticity label. In this case, the certificate identification model can still output the basis for the determination corresponding to the certificate authenticity determination result to improve the interpretability and credibility of the model. Specifically, the basis for the determination can be automatically generated by the model based on the input certificate image and the features of the certificate anti-counterfeiting rules, without manual labeling.

[0056] Exemplary implementation methods include, but are not limited to, the following: ① Attention weight mapping: when processing the certificate image and the anti-counterfeiting rule features, the model calculates the contribution of each feature to the final authenticity determination, and the information corresponding to the features with higher contribution is taken as the basis for the determination. For example, the model can identify that the features of a certain watermark region are abnormal, or the certificate number coding deviates from the pre-set rules, thereby generating the corresponding basis for the determination.

[0057] ② Feature difference analysis: the model calculates the matching degree between the image features and the rule features, and when the matching degree of certain features deviates significantly from the normal range, these features are taken as the basis for the determination to explain the determination result.

[0058] ③ Generative explanation module: the model can integrate a small generative module to automatically convert the analysis results of the image and rule features into text form of the basis for the determination, such as "watermark position deviates from the rule region" or "certificate number coding does not meet the specification".

[0059] To ensure the accuracy of the decision basis, the following measures can be taken: ① Indirectly against the real label: combine the model-generated decision basis with the model's true or false prediction results to verify the rationality of the decision basis on the correctly predicted samples. ② Expert or rule verification: On some samples, artificial or rule engine can verify the decision basis generated by the model to evaluate its consistency with the actual anti-counterfeiting rules. ③ Confidence scoring mechanism: the model outputs a confidence score for each decision basis, combined with the overall prediction result, to improve the reliability of the decision basis.

[0060] Through the above implementation, even without human-annotated decision basis, the model can still generate explanatory information, making the certificate authenticity judgment result more transparent. This way not only improves the credibility of the model in practical application, but also enhances its adaptability and generalization performance to unknown certificate types.

[0061] In some embodiments, in order to enhance the attention ability of the certificate identification model to the key anti-counterfeiting features of the certificate, the certificate image can be processed so that the visual saliency of the position of the anti-counterfeiting element in the certificate image is higher than that of other positions. In this way, the model can more effectively learn the features of the anti-counterfeiting elements during the training process, improving the recognition ability of the certificate authenticity. Please refer to Figure 3 , which shows the original certificate image and the processed certificate image.

[0062] For example, the visual saliency of the position of the anti-counterfeiting element in the certificate image is higher than that of other positions, including at least one of the following: ① Transparency enhancement method: the transparency of the position of the anti-counterfeiting element in the certificate image is higher than that of other positions. For example, a Mask (masking image) with the same size as the certificate image can be generated, which has a higher transparency, such as semi-transparency, at the position of the anti-counterfeiting element, and a lower transparency at other positions. Then, the Mask is superimposed on the certificate image, so that the transparency of the position of the anti-counterfeiting element in the superimposed certificate image is higher than that of other positions. Through this processing method, the model can assign higher attention weight to the area where the anti-counterfeiting element is located during training, so as to more accurately extract the anti-counterfeiting features.

[0063] ② Brightness enhancement method: the brightness of the position of the anti-counterfeiting element in the certificate image is higher than that of other positions. The brightness of the position of the anti-counterfeiting element can be set higher than that of other positions by adjusting the image brightness. For example, for key anti-counterfeiting areas such as watermarks or security lines, local brightness values can be added to the original image to make these areas more visually prominent. This method can guide the model to focus more on the anti-counterfeiting elements with high brightness in the convolution feature extraction or attention mechanism, thereby improving the feature extraction efficiency.

[0064] ③Color saturation enhancement: The color saturation of the location where the anti-counterfeiting element in the certificate image is higher than that of other locations. By adjusting the color saturation of the anti-counterfeiting element area, it can be higher than the saturation of other areas of the image. For example, for color anti-counterfeiting patterns, special fonts or anti-counterfeiting graphics area, the color saturation can be enhanced to make it more visually prominent. This processing method can assist the model in focusing on key anti-counterfeiting elements during multi-channel image feature extraction, improving the discrimination ability.

[0065] ④Edge enhancement: The edge contrast of the location where the anti-counterfeiting element in the certificate image is higher than that of other locations. By enhancing the edge contrast of the anti-counterfeiting element area, the boundary of the watermark, security line, special pattern, etc. is more clear. The implementation method includes using a convolution filter to extract high-frequency edge information in the anti-counterfeiting element area, and then superimposing it with the original image to enhance the local structure features. The model can more easily capture the shape and structure features of the anti-counterfeiting element during training or inference.

[0066] In the present embodiment, the above-mentioned visual saliency enhancement methods can be used alone or in combination. By performing visual saliency enhancement processing on the certificate image, the certificate recognition model can more effectively learn the spatial distribution features and surface features of the anti-counterfeiting elements during the training process, thereby improving the accuracy and robustness of the certificate authenticity identification, especially when facing high-imitation counterfeit certificates.

[0067] For example, the certificate image contains at least one anti-counterfeiting element to assist the certificate recognition model in identifying the authenticity of the certificate. In order to enhance the attention ability of the model to the key anti-counterfeiting elements, the visual saliency of the location where different anti-counterfeiting elements are located can be set.

[0068] In one possible implementation, the visual saliency of the location where different anti-counterfeiting elements in the certificate image are located can be set to be the same, that is, all anti-counterfeiting elements are equally prominent in vision. This method is suitable for the case where the contribution of anti-counterfeiting elements to authenticity judgment is roughly equivalent, and can make the model pay equal attention to all anti-counterfeiting elements and improve the model's perception ability of the overall anti-counterfeiting features.

[0069] In another possible implementation, the visual saliency degree of the position where each anti-counterfeiting element is located can be set differently according to the importance or priority of the anti-counterfeiting element. Specifically, the visual saliency degree of the position where any anti-counterfeiting element is located is positively correlated with the priority of the anti-counterfeiting element, that is, the higher the priority of the anti-counterfeiting element, the higher the visual saliency degree. For example, for an ID card, the watermark can have a high priority, and the paper texture feature can have a relatively low priority, and thus the visual saliency degree of the position where the watermark is located can be higher than that of the position where the paper texture feature is located. By setting differently, the model can pay more attention to the key anti-counterfeiting elements that contribute more to the judgment of the authenticity of the certificate during the training process, thereby improving the accuracy and robustness of the model in identifying high-imitation counterfeit certificates or certificates with complex anti-counterfeiting designs. At the same time, this setting method can be combined with the visual saliency enhancement methods such as transparency, brightness, saturation, or edge to realize the key learning of multi-dimensional features.

[0070] After the certificate identification model is trained, the model can be deployed on a server, a client, or a mobile terminal, to realize the authenticity identification of an input certificate image. In this stage, the model can receive a certificate image photographed or uploaded by a user, and corresponding certificate anti-counterfeiting rule information, and make an authenticity judgment on the certificate through the association between the image features and the anti-counterfeiting rule features learned in the foregoing training process.

[0071] Please refer to Figure 4 The embodiments of the present specification provide a certificate identification method, which can be executed by an electronic device, including but not limited to a server (such as a physical server, a virtual server, etc.), a smart phone / mobile phone, a tablet computer, a personal digital assistant (PDA), a laptop computer, a desktop computer, a wearable device, or other types of devices. The certificate identification method includes: In S400, a certificate image to be identified is obtained, and a certificate type to which the certificate image to be identified belongs is determined.

[0072] Exemplarily, the determination manner of the certificate type to which the certificate image to be identified belongs includes but is not limited to at least one of the following: (1) the user can directly input the certificate type information, such as "Chinese ID card", "passport", "driver's license", etc., and the electronic device can directly use the certificate type information input by the user, or the electronic device can match the user input with a preset certificate type list to determine the standardized certificate type information. (2) the electronic device can use a trained certificate type identification model to analyze the certificate image, so as to automatically identify the certificate type. The certificate type identification model can be a convolutional neural network (CNN), a visual Transformer (ViT), or a multi-modal large model, which can extract overall layout, text layout, pattern features, etc. of the certificate image for classification. The probability distribution output by the model can be used to select the certificate type corresponding to the maximum probability, or a confidence threshold can be used for classification decision. (3) the user input certificate type information and the prediction result of the certificate type identification model are combined to make a comprehensive judgment, so as to improve the accuracy and robustness of the certificate type identification. For example, when the user input certificate type is consistent with the model prediction type, the type is directly adopted; if not, the confidence or priority rule can be used for adjustment.

[0073] In S402, based on the certificate image to be identified and / or the certificate type thereof, a matched certificate anti-counterfeiting rule is obtained by querying from a certificate anti-counterfeiting rule library.

[0074] Exemplarily, the query manner of the certificate anti-counterfeiting rule can include but is not limited to: (1) certificate type-based retrieval: directly positioning the corresponding anti-counterfeiting rule item in the rule library through the standardized certificate type information, to obtain the watermark position and features, security line features, anti-counterfeiting patterns, special font layout, paper material, coding rules, etc. (2) image-based retrieval: using the certificate image features to generate a query vector, and performing similarity matching in the anti-counterfeiting rule library to find the most matched anti-counterfeiting rule of the input certificate. (3) hybrid retrieval manner: combined certificate type information and image features for joint retrieval, first narrowing down the candidate rule set through the certificate type, and then performing fine matching based on the image features to improve the accuracy and efficiency of the query.

[0075] In S404, the certificate image to be identified and the certificate anti-counterfeiting rule are input into the trained certificate identification model, and a certificate identification model output certificate authenticity judgment result is obtained.

[0076] In this embodiment, the certificate image to be identified and the certificate anti-counterfeiting rule are input into the trained certificate identification model, and the certificate identification model analyzes the input image features and anti-counterfeiting rule features to output a certificate authenticity judgment result.

[0077] Through the above embodiments, the electronic device can realize automatic recognition and authenticity judgment of multiple types of certificates, avoid the limitations of manual inspection, and improve processing efficiency and recognition accuracy. The method can be extended to various types of certificates worldwide, further improving the discrimination ability and robustness of the model.

[0078] In some embodiments, the electronic device can also use the fake certificate image library to assist the identification process during certificate authenticity identification, thereby further improving the discrimination ability of the certificate identification model in the high-imitation fake certificate scenario. Specifically, the electronic device first calculates the similarity between the to-be-identified certificate image and each fake certificate image in the fake certificate image library. The similarity calculation can be based on the cosine similarity, Euclidean distance, or other feature matching algorithms of the image feature vector, and the image features can be extracted by a convolutional neural network, a visual Transformer, or a multi-modal large model.

[0079] After obtaining the similarity result, the electronic device can filter out at least one fake certificate image similar to the to-be-identified certificate image from the fake certificate image library based on a set similarity filtering rule, such as filtering out the fake certificate image corresponding to the similarity greater than the set similarity threshold, or filtering out the N fake certificate images with the largest similarity. By introducing fake samples similar to the to-be-identified certificate, the sensitivity of the certificate identification model to high-imitation fake features can be enhanced, and the discrimination ability and robustness of the model in complex fake scenarios can be improved.

[0080] Subsequently, please refer to Figure 5 , the electronic device inputs the to-be-identified certificate image, the corresponding certificate anti-counterfeiting rule, and the filtered fake certificate image into the trained certificate identification model. The certificate identification model integrates the image features of the to-be-identified certificate, the anti-counterfeiting rule features, and the fake certificate features to make a comprehensive judgment on the certificate authenticity, and outputs the certificate authenticity judgment result.

[0081] In some embodiments, the fake certificate image library can be continuously updated, including high-imitation fake certificate images collected by humans or generated by models, to adapt to changing counterfeiting methods. By introducing fake certificate images similar to the to-be-identified certificate image, combined with multi-modal feature fusion and anti-counterfeiting rule assisted judgment, the certificate identification method of the present application can significantly improve the recognition accuracy of high-imitation fake certificates, meet the needs of global diversified certificate verification, and enhance the security and reliability of the system.

[0082] In one possible implementation, please refer to Figure 5The certificate identification model can also output the basis for the certificate authenticity judgment result, such as "watermark position inconsistent", "security line missing", "font error", "similar to the known fake certificate forgery method", and the like, to enhance the interpretability and credibility of the certificate authenticity judgment result.

[0083] In one possible implementation, referring to Figure 5 The certificate identification model can also output a credibility score representing the confidence level of the model on the judgment result.

[0084] The various technical features in the above embodiments can be combined in any manner, as long as the combination of features does not conflict or contradict. However, due to the limited space, not all possible combinations are described, and any combination of the various technical features in the above embodiments is within the scope of the present disclosure.

[0085] Figure 6 is a schematic structural diagram of a device provided by an exemplary embodiment. As shown in Figure 6 The device 600 mainly consists of a communication interface 602, a user interface 604, a processor 606, and a data storage 608, which are connected and communicate with each other through a system bus, a network, or other connection mechanism 610. The communication interface 602 enables the device 600 to communicate with other devices, access networks, and transmission networks through analog or digital modulation. For example, the communication interface 602 can include a chipset and an antenna for wireless communication with a radio access network or an access point. In addition, the communication interface 602 can also be a wired interface such as an Ethernet, token ring, or USB port, or a wireless interface such as Wifi, Bluetooth, Global Positioning System (GPS), or a wide area wireless interface (e.g., WiMAX or LTE). Of course, the communication interface 602 can also support other forms of physical layer interfaces and standard or proprietary communication protocols. The communication interface 602 can also include multiple physical communication interfaces, such as a Wifi interface, a Bluetooth interface, and a wide area wireless interface.

[0086] The user interface 604 includes input components such as a keypad, keyboard, touch- sensitive or presence-sensitive panel, computer mouse, trackball, joystick, microphone, still camera, and video camera, among others, and output components such as a display screen (which can be combined with a touch-sensitive panel), a CRT, LCD, LED, display using DLP technology, printer, other known or future developed equivalent devices, among others. The user interface 604 can also generate auditory output through a speaker, a speaker jack, an audio output port, an audio output device, earphones, and other known or future developed equivalent devices. In some embodiments, the user interface 604 can include software, circuitry, or other forms of logic that enables the device 600 to transmit data to and receive data from external user input / output devices. In addition or alternatively, the device 600 can support remote access from other devices through the communication interface 602 or another physical interface (not shown). The user interface 604 can be configured to receive user input, the location and movement of which can be indicated by a pointer or cursor as described herein. The user interface 604 can also be configured as a display device for rendering or displaying a text snippet.

[0087] The processor 606 can include one or more general-purpose processors and / or special-purpose processors.

[0088] The data storage 608 can include one or more volatile and / or non-volatile storage components, and can be integrated in whole or in part with the processor 606. The data storage 608 can include removable and / or non-removable components.

[0089] The processor 606 is capable of executing program instructions 618 (e.g., compiled or interpreted program logic and / or machine code) stored in the data storage 608 to implement various functionality described herein. The data storage 608 can include non-transitory computer-readable media having stored thereon program instructions that, when executed by the device 600, enable the device 600 to carry out any of the methods, processes, or functions disclosed in the specification and / or drawings. Execution of the program instructions 618 by the processor 606 can cause the processor 606 to utilize data 612.

[0090] For example, the program instructions 618 can include an operating system 622 (e.g., an operating system kernel, device drivers, and / or other modules) installed on the device 600, as well as one or more application programs 620 (e.g., a browser, a social application, or a gaming application). Similarly, the data 612 can include operating system data 614 and application data 616. The operating system data 614 is primarily accessible to the operating system 622, while the application data 616 is primarily accessible to the one or more application programs 620. The application data 616 can be located in a file system that is visible or hidden to a user of the device 600.

[0091] The application programs 620 can communicate with the operating system 622 through one or more application programming interfaces (APIs). These APIs can help the application programs 620 read and / or write application data 616, transfer or receive information via the communication interface 602, receive or display information on the user interface 604, etc.

[0092] In some terminology, the application programs 620 can be referred to simply as "apps." In addition, the application programs 620 can be downloaded to the device 600 through one or more online application stores or application markets. However, the application programs can also be installed on the device 600 through other means, such as through a web browser or a physical interface on the device 600 (e.g., a USB port).

[0093] In some embodiments, the certificate identification model training apparatus can be applied in a device as shown in FIG. 1 to implement the technical solutions of the present specification. The certificate identification model training apparatus can include: Figure 6 a sample obtaining module, configured to obtain a certificate sample set mixed with multiple certificate types, the certificate sample set including multi-modal certificate samples labeled with certificate authenticity labels, the multi-modal certificate samples including certificate images and certificate anti-counterfeiting rules, and the certificate anti-counterfeiting rules of any certificate type being used to describe the positions of at least one anti-counterfeiting element of the certificate type in the certificate and the anti-counterfeiting rules of the anti-counterfeiting element; a training module, configured to input the multi-modal certificate samples into a certificate identification model to be trained, so that the certificate identification model to be trained performs identification based on the certificate images and the certificate anti-counterfeiting rules, and outputs certificate authenticity judgment results; The training module is further configured to train the certificate identification model with the optimization target of minimizing errors between the certificate authenticity labels and the certificate authenticity judgment results.

[0094] In an implementation manner, the multi-modal certificate samples include positive samples and negative samples; the certificate images and the certificate anti-counterfeiting rules in the positive samples have a matching relationship, and the certificate images and the certificate anti-counterfeiting rules in the negative samples do not have a matching relationship.

[0095] The optimization target further includes at least one of the following: minimizing distances between features extracted by the certificate identification model from the certificate images and the certificate anti-counterfeiting rules in the positive samples, and maximizing distances between features extracted by the certificate identification model from the certificate images and the certificate anti-counterfeiting rules in the negative samples.

[0096] ​The positive sample includes a certificate image belonging to any certificate type and a correct anti-counterfeiting rule corresponding to the certificate type; wherein the correct anti-counterfeiting rule is obtained from a certificate anti-counterfeiting rule library based on the certificate type to which the certificate image belongs, and the certificate anti-counterfeiting rule library involves more types of certificates than the types of certificates involved in model training.

[0097] The negative sample includes at least one of the following: a certificate image belonging to any certificate type and an incorrect anti-counterfeiting rule corresponding to the certificate type, a certificate image belonging to a second certificate type and a correct anti-counterfeiting rule corresponding to a first certificate type, and a certificate image belonging to the first certificate type and a correct anti-counterfeiting rule corresponding to the second certificate type.

[0098] In an implementation manner, the visual saliency degree of the position of the anti-counterfeiting element in the certificate image is higher than that of other positions.

[0099] The certificate image contains at least one anti-counterfeiting element; the visual saliency degrees of the positions of different anti-counterfeiting elements are the same; or the visual saliency degrees of the positions of different anti-counterfeiting elements are set differently based on the priorities of the different anti-counterfeiting elements, and the visual saliency degree of the position of any anti-counterfeiting element is positively correlated with the priority of the anti-counterfeiting element.

[0100] The visual saliency degree of the position of the anti-counterfeiting element in the certificate image is higher than that of other positions, including at least one of the following: the transparency of the position of the anti-counterfeiting element in the certificate image is higher than that of other positions; the brightness of the position of the anti-counterfeiting element in the certificate image is higher than that of other positions; the color saturation of the position of the anti-counterfeiting element in the certificate image is higher than that of other positions; and the edge contrast of the position of the anti-counterfeiting element in the certificate image is higher than that of other positions.

[0101] In some embodiments, the certificate identification device can be applied to the device as shown in Figure 6 The certificate identification device can include: A certificate image acquisition module is configured to acquire a certificate image to be identified and determine a certificate type to which the certificate image to be identified belongs.

[0102] A certificate anti-counterfeiting rule query module is configured to query a matched certificate anti-counterfeiting rule from a certificate anti-counterfeiting rule library based on the certificate image to be identified and / or the certificate type to which the certificate image to be identified belongs.

[0103] A model inference module is configured to input the certificate image to be identified and the certificate anti-counterfeiting rule into a trained certificate identification model to obtain a certificate authenticity judgment result output by the certificate identification model.

[0104] In an implementation manner, the device further comprises a fake certificate image screening module, configured to calculate similarity between the to-be-identified certificate image and fake certificate images in a fake certificate image library; and screen at least one fake certificate image from the fake certificate image library based on the similarity between the to-be-identified certificate image and the fake certificate images in the fake certificate image library.

[0105] The model inference module is specifically configured to input the to-be-identified certificate image, the certificate anti-counterfeiting rule, and the screened fake certificate image into a trained certificate identification model.

[0106] For the convenience of description, the above device is described as various modules or units in function. Of course, the functions of each module or unit can be implemented in the same or more software and / or hardware when implementing one or more of the present specification, and the modules implementing the same function can be implemented by a combination of multiple sub-modules or sub-units. The above-described device embodiments are only illustrative, for example, the division of the units is only a logical function division, and actual implementation can have another division manner, for example, multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed.

[0107] Based on the same idea as the above method, the present specification also provides an electronic device, comprising: a processor; a memory for storing processor executable instructions; wherein the processor implements the steps of the method according to any one of the above embodiments by running the executable instructions.

[0108] Based on the same idea as the above method, the present specification also provides a computer readable storage medium, which stores computer instructions, and the instructions are executed by a processor to implement the steps of the method according to any one of the above embodiments.

[0109] Based on the same idea as the above method, the present specification also provides a computer program product, comprising computer program / instructions, which are executed by a processor to implement the steps of the method according to any one of the above embodiments.

[0110] Those skilled in the art can understand that: In the present specification, the term "comprise", "include" or any other variant thereof is intended to cover non-exclusive inclusion, so that a process, method, product or device including a series of elements not only includes those elements, but also includes other elements not explicitly listed or inherent to such a process, method, product or device. Without more limitations, it does not exclude the presence of other same or equivalent elements in the process, method, product or device including the elements.

[0111] In this specification, "one", "a", and "the" do not refer to the singular, but can include the plural as well, unless explicitly stated otherwise.

[0112] In this specification, the ordinal numbers first, second, etc. do not necessarily indicate the order or sequence, and in many cases are used to distinguish between objects. For example, the first server and the second server usually refer to two servers. In order to distinguish between the two servers, they are expressed as the first server and the second server. Of course, at times, the two servers can also be the same server.

[0113] In this specification, unless specifically stated otherwise, "receiving and sending data" is not necessarily direct receiving and sending, and can be indirect receiving and sending. For example, A receiving data sent by B can be understood as A directly receiving data sent by B, or can be understood as A indirectly receiving data sent by B through C or other subjects. Similarly, B sending data to A can be understood as B directly sending data to A, or can be understood as B indirectly sending data to A through C or other subjects. Here, C can be one subject, or two or more subjects.

[0114] In this specification, unless specifically stated otherwise, the association relationship generated between structures can be a direct association relationship or an indirect association relationship. For example, when describing "A is connected with B", unless it is specifically stated that A is directly connected with B, it should be understood that A can be directly connected with B, or A can be indirectly connected with B; for another example, when describing "A is on B", unless it is specifically stated that A is directly above B (AB is adjacent and A is above B), it should be understood that A can be directly above B, or A can be indirectly above B (there are other elements between AB, and A is above B). By analogy.

[0115] This specification uses specific words to describe the embodiments of this specification. As "one embodiment", "an embodiment", and / or "some embodiments" means a certain feature, structure or characteristic related to at least one embodiment of this specification. Therefore, it should be emphasized and noted that the "an embodiment" or "one embodiment" or "one alternative embodiment" mentioned in different places in this specification does not necessarily refer to the same embodiment. In addition, the skilled in the art can combine and combine the different embodiments or examples described in this specification and the features of different embodiments or examples, without contradiction.

[0116] While one or more embodiments of the specification provide method steps as described in the embodiments or flowcharts, it is understood that the order of steps recited in the embodiments or flowcharts is merely one of many possible sequences of execution of the steps, and that variations in the order of execution of the steps, or parallel execution of the steps, are within the scope of the claims when the claims refer to the steps.

Claims

1. A method for training a certificate identification model, comprising: obtaining a certificate sample set of mixed multiple certificate types, wherein the certificate sample set comprises multi-modal certificate samples labeled with certificate authenticity labels, and the multi-modal certificate samples comprise certificate images and certificate anti-counterfeiting rules, wherein the certificate anti-counterfeiting rules of any certificate type are used to describe the positions of at least one anti-counterfeiting element of the certificate type in the certificate and the anti-counterfeiting rules of the anti-counterfeiting element; inputting the multi-modal certificate samples into a certificate identification model to be trained, so as to output certificate authenticity judgment results based on the certificate images and the certificate anti-counterfeiting rules by the certificate identification model to be trained; training the certificate identification model by minimizing the error between the certificate authenticity labels and the certificate authenticity judgment results.

2. The method of claim 1, wherein the multi-modal certificate samples comprise positive samples and negative samples; the certificate images and the certificate anti-counterfeiting rules in the positive samples have a matching relationship, and the certificate images and the certificate anti-counterfeiting rules in the negative samples do not have a matching relationship; the optimization target further comprises at least one of the following: minimizing the distance between the features extracted from the certificate images and the certificate anti-counterfeiting rules in the positive samples by the certificate identification model, and maximizing the distance between the features extracted from the certificate images and the certificate anti-counterfeiting rules in the negative samples by the certificate identification model.

3. The method of claim 2, The positive sample includes a certificate image belonging to any certificate type and a correct anti-counterfeiting rule corresponding to the certificate type; wherein the correct anti-counterfeiting rules are obtained by querying a certificate anti-counterfeiting rule library based on the certificate type of the certificate image, and the certificate anti-counterfeiting rule library involves more types of certificates than the types of certificates involved in the model training; the negative samples comprise at least one of the following: a certificate image belonging to any certificate type and an incorrect anti-counterfeiting rule corresponding to the certificate type, a certificate image belonging to a second certificate type and a correct anti-counterfeiting rule corresponding to a first certificate type, and a certificate image belonging to the first certificate type and a correct anti-counterfeiting rule corresponding to the second certificate type.

4. The method of claim 1, wherein the visual saliency of the position of the anti-counterfeiting element in the certificate image is higher than that of other positions; wherein, the certificate image contains at least one anti-counterfeiting element; the visual saliency of the positions of different anti-counterfeiting elements is the same; or the visual saliency of the positions of different anti-counterfeiting elements is set differently based on the priority of the different anti-counterfeiting elements, and the visual saliency of the position of any anti-counterfeiting element is positively correlated with the priority of the anti-counterfeiting element.

5. The method of claim 4, wherein the visual saliency of the position of the anti-counterfeiting element in the certificate image is higher than that of other positions, comprising at least one of the following: the transparency of the position of the anti-counterfeiting element in the certificate image is higher than that of other positions; the brightness of the position of the anti-counterfeiting element in the certificate image is higher than that of other positions; the color saturation of the position of the anti-counterfeiting element in the certificate image is higher than that of other positions; and the edge contrast of the position of the anti-counterfeiting element in the certificate image is higher than that of other positions.

6. A method for identification of a certificate, comprising: obtaining a certificate image to be identified, and determining a type of certificate to which the certificate image to be identified belongs; querying a matched certificate anti-counterfeiting rule from a certificate anti-counterfeiting rule library based on the certificate image to be identified and / or the type of certificate to which the certificate image to be identified belongs; inputting the certificate image to be identified and the certificate anti-counterfeiting rule into a trained certificate identification model to obtain a certificate authenticity judgment result output by the certificate identification model, wherein the certificate identification model is trained based on the method of any one of claims 1 to 5.

7. The method of claim 6, further comprising: calculating a similarity between the certificate image to be identified and a counterfeit certificate image in a counterfeit certificate image library; based on the similarity between the certificate image to be identified and the counterfeit certificate image in the counterfeit certificate image library, screening at least one counterfeit certificate image from the counterfeit certificate image library; the inputting the certificate image to be identified and the certificate anti-counterfeiting rule into the trained certificate identification model comprises: inputting the certificate image to be identified, the certificate anti-counterfeiting rule, and the screened counterfeit certificate image into the trained certificate identification model.

8. An electronic device, comprising: comprising: a processor; a memory for storing processor-executable instructions, wherein the processor implements the steps of the method of any one of claims 1 to 7 by running the executable instructions.

9. A computer-readable storage medium, characterized in that, a computer program / product / instructions stored thereon, which, when executed by a processor, implement the steps of the method of any one of claims 1 to 7.

10. A computer program product, characterised in that, a computer program / product / instructions stored thereon, which, when executed by a processor, implement the steps of the method of any one of claims 1 to 7.

Images