Artwork authentication method based on image recognition

Abstract

The application discloses an art work identification method based on image recognition and relates to the technical field of art work identification.The method comprises the following steps: acquiring an art work image, preprocessing the art work image, and extracting a feature vector of the preprocessed art work image; performing reverse image search, calculating the similarity between an art work image uploaded by a user and an art work image in a database, and performing image comparison; and constructing an art work identification model to identify the authenticity of the art work.The application collects multi-source art work images and extracts feature vectors, integrates the image features uploaded by the user with the image features with the highest similarity in the database, inputs the final features obtained through graph convolution operation and deep neural network processing into the constructed art work identification model to identify the authenticity of the art work, enhances the capturing ability of art work details, can deeply extract the image features of the art work and effectively distinguish genuine products from fake products, and improves the identification accuracy.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of art work identification, in particular to an art work identification method based on image recognition. BACKGROUND

[0002] The prosperity of art work market has driven the demand for art work identification methods. Traditional art work identification relies on the experience and intuitive judgment of experts. This method is not only time-consuming and labor-intensive, but also susceptible to subjective factors, with limited accuracy and objectivity. In recent years, with the rapid development of computer vision and machine learning technology, art work identification methods based on image recognition have gradually emerged. Such technology analyzes the digital images of art works to identify their style, brush strokes, and materials used, thereby determining the authenticity of art works. Although existing image recognition technology has been applied in many fields, it still faces unique challenges in the field of art work identification. Factors such as image quality, lighting conditions, and shooting angles of art works can affect the accuracy of identification results. Existing technology relies on the analysis of global image features, but lacks the ability to identify local features of art works, making it difficult to accurately identify extremely similar or slightly different art works. Existing automatic identification technology also fails to fully utilize user input and historical data, limiting the adaptability and flexibility of the identification method. SUMMARY

[0003] In view of the above-mentioned problems existing in the prior art of art work identification based on image recognition, the present application is proposed.

[0004] Therefore, the problem to be solved by the present application is that existing technology relies on the analysis of global image features, but lacks the ability to identify local features of art works, making it difficult to accurately identify extremely similar or slightly different art works. Existing automatic identification technology also fails to fully utilize user input and historical data, limiting the adaptability and flexibility of the identification method.

[0005] To solve the above technical problems, the present application provides the following technical solution: an art work identification method based on image recognition, which includes obtaining an art work image and preprocessing the art work image to extract a feature vector of the preprocessed art work image; performing reverse image search and calculating the similarity between a user-uploaded art work image and art work images in a database for image comparison; constructing an art work identification model for art work authenticity identification; and labeling the identified art work, securely storing the data, and implementing access control.

[0006] As a preferred scheme of the image recognition-based art work identification method, the method comprises the following steps: obtaining an art work image and preprocessing the art work image, wherein the step of obtaining the art work image and preprocessing the art work image comprises the following steps: obtaining a true or false art work image from various data sources, taking three images from the front, side and back of the art work, performing color correction on the collected images, adjusting the color temperature and color saturation of the images, enhancing the contrast of the images by using adaptive histogram equalization, removing image noise by using a Gaussian filter, adjusting the size of the denoised image to 299*299 pixels, normalizing the image pixel value to the range of 0-1, assigning a unique serial number to the preprocessed art work image and marking the true or false information and artist information of the art work, and storing the preprocessed image data set in a database.

[0007] As a preferred scheme of the image recognition-based art work identification method, the method comprises the following steps: obtaining an art work image and preprocessing the art work image, wherein the step of obtaining the art work image and preprocessing the art work image comprises the following steps: obtaining a true or false art work image from various data sources, taking three images from the front, side and back of the art work, performing color correction on the collected images, adjusting the color temperature and color saturation of the images, enhancing the contrast of the images by using adaptive histogram equalization, removing image noise by using a Gaussian filter, adjusting the size of the denoised image to 299*299 pixels, normalizing the image pixel value to the range of 0-1, assigning a unique serial number to the preprocessed art work image and marking the true or false information and artist information of the art work, and storing the preprocessed image data set in a database.

[0008] As a preferred scheme of the image recognition-based art work identification method, the method comprises the following steps: obtaining an art work image and preprocessing the art work image, wherein the step of obtaining the art work image and preprocessing the art work image comprises the following steps: obtaining a true or false art work image from various data sources, taking three images from the front, side and back of the art work, performing color correction on the collected images, adjusting the color temperature and color saturation of the images, enhancing the contrast of the images by using adaptive histogram equalization, removing image noise by using a Gaussian filter, adjusting the size of the denoised image to 299*299 pixels, normalizing the image pixel value to the range of 0-1, assigning a unique serial number to the preprocessed art work image and marking the true or false information and artist information of the art work, and storing the preprocessed image data set in a database.

[0009] As a preferred scheme of the image recognition-based art work identification method, the method comprises the following steps: obtaining an art work image and preprocessing the art work image, wherein the step of obtaining the art work image and preprocessing the art work image comprises the following steps: obtaining a true or false art work image from various data sources, taking three images from the front, side and back of the art work, performing color correction on the collected images, adjusting the color temperature and color saturation of the images, enhancing the contrast of the images by using adaptive histogram equalization, removing image noise by using a Gaussian filter, adjusting the size of the denoised image to 299*299 pixels, normalizing the image pixel value to the range of 0-1, assigning a unique serial number to the preprocessed art work image and marking the true or false information and artist information of the art work, and storing the preprocessed image data set in a database.

[0010]

[0011] Wherein sim(A, B) is the similarity between the user-uploaded art image feature vector A and the art image feature vector B in the database, I(A) is the interest vector of the user-uploaded art image feature vector A, I(B) is the interest vector of the art image feature vector B in the database, ||I(A)|| and ||I(B)|| are the L2 norms of the interest vectors of the user-uploaded art image feature vector A and the art image feature vector B in the database respectively;

[0012] The art image is sorted in descending order according to the similarity score, the art image with the highest similarity is selected, and a list of similar art images is displayed according to the similarity score, including the image, artist information and similarity score of each art image;

[0013] The extracted user-uploaded art image feature vector A and the art image feature vector B in the database are combined into a feature vector F as follows:

[0014] F = [A, B].

[0015] As a preferred scheme of the art identification method based on image recognition, wherein the art identification model comprises,

[0016] An art relationship graph is constructed, nodes in the graph structure are defined, each node represents an art, the node feature is the feature vector F, edges in the graph structure are defined, and the edges represent the similarity relationship between the art, and a graph G containing all art and the similarity relationship is constructed;

[0017] The art relationship graph is subjected to graph convolution operation, and the initial feature vector H 0 of each node is initialized as the node feature F, multi-layer graph convolution is performed, and the feature of each node i is subjected to graph convolution update:

[0018]

[0019] In the formula, is the feature representation of node i in the l+1 layer, sigma is the ReLU activation function, N(i) is the neighbor node set of node i, d i and d j are the neighbor node numbers of node i and node j respectively, W (l) is the weight matrix of the l layer, is the feature representation of neighbor node j in the l layer, b (l) is the bias term of the l layer;

[0020] After multi-layer graph convolution, the final feature representation M of each node is obtained. The final feature representation M of all nodes is collected and input into the fully connected layer of the deep neural network to further extract high-level feature representations, resulting in the feature representation M′ after processing by the deep neural network.

[0021] Building an art identification model based on support vector machines:

[0022] D = sign(ω*M′+b),

[0023] Where D is the model output, M′ is the feature representation after processing by the deep neural network, ω is the model weight, b is the bias term, sign is the classification function, +1 represents real and -1 represents fake;

[0024] The image dataset stored in the database is used as the model training set. The model is trained using the training set, and the model optimization objective is defined as follows:

[0025]

[0026] Where Dq is the output of the model on the q-th training data, M′q is the output of the q-th training data, and subject defines the optimization constraints. This means that all q constraints are true;

[0027] The model parameters are obtained by solving the optimization objective, and then input into the model to obtain the artwork identification model.

[0028] As a preferred embodiment of the image recognition-based artwork identification method of the present invention, the artwork authenticity identification refers to combining the feature vector of the artwork image uploaded by the user with the feature vector of the artwork image with the highest similarity in the database, and obtaining the final feature representation M′ after graph convolution operation and deep neural network processing, and inputting the obtained final feature representation M′ into the artwork identification model to obtain the artwork identification result.

[0029] In a preferred embodiment of the image recognition-based artwork authentication method of the present invention, the step of labeling the artwork after authentication refers to labeling the artwork based on the authentication results.

[0030] If an artwork is identified as a genuine artwork, it will be labeled as "authentic" and a digital certificate will be generated for each artwork identified as authentic, including the artist's information, the artwork's historical ownership, and its sale and exhibition history.

[0031] If the artistic work is identified as a fake artistic work, the fake artistic work is marked with a "fake" label, the record includes the identification date and the confidence, and the reason for identifying as fake is provided.

[0032] As a preferred solution of the image recognition-based artistic work identification method, the data is securely stored and access control is implemented by storing the information of the artistic work, the identification result and the timestamp in a database, and setting security access measures, the database stores the data for cloud backup, and periodically checks the integrity of the stored data and the backup data, and generates an integrity detection record after the detection is completed, which is stored in the database, and allows authorized users to query the artistic work identification result through a user interface.

[0033] A computer device comprises a memory and a processor, and the memory stores a computer program, characterized in that the processor executes the computer program to realize the steps of the image recognition-based artistic work identification method.

[0034] A computer readable storage medium stores a computer program, characterized in that the computer program is executed by a processor to realize the steps of the image recognition-based artistic work identification method.

[0035] The present application has the following advantages: the present application collects multi-source artistic work images and extracts feature vectors, integrates the image features uploaded by the user with the highest similarity image features in the database, and inputs the final features obtained through graph convolution operation and deep neural network processing into the constructed artistic work identification model for artistic work authenticity identification, enhances the capturing ability of artistic work details, can deeply extract image features of artistic works and effectively distinguish genuine and fake products, and improves the accuracy of identification. BRIEF DESCRIPTION OF DRAWINGS

[0036] In order to more clearly illustrate the technical solutions of the embodiments of the present application, the following will briefly introduce the drawings needed in the embodiment description. Obviously, the drawings in the following description are only some embodiments of the present application, and those skilled in the art can obtain other drawings according to these drawings without creative labor.

[0037] Figure 1 The flowchart of the image recognition-based artistic work identification method.

[0038] Figure 2 The implementation schematic diagram of the image recognition-based artistic work identification method. DETAILED DESCRIPTION

[0039] In order to make the above objectives, features and advantages of the present application more apparent, specific embodiments of the present application will be described in detail below with reference to the accompanying drawings.

[0040] In the following description, numerous specific details are set forth in order to provide a thorough understanding of the present application. The present application, however, can be practiced in a variety of ways other than those specifically described herein, and the scope of the present application is not limited to the embodiments described herein. It can be appreciated that the specific structural and functional details disclosed herein are merely representative in terms of their

[0041] Secondly, the term "one embodiment" or "an embodiment" as used herein means that a particular implementation can include a particular feature, structure, or characteristic, but every embodiment can not necessarily include the particular feature, structure, or characteristic. Moreover, such features, structures, or characteristics can be combined in one or more implementations in any suitable manner. In the following description, numerous specific details are set forth in order to provide a thorough understanding of the present application. The present application, however, can be practiced in a variety of ways other than those specifically described herein, and the scope of the present application is not limited to the embodiments described herein. It can be appreciated that the specific structural and functional details disclosed herein are merely representative in terms of their

[0042] Embodiment 1

[0043] Reference Figure 1 and Figure 2 For the first embodiment of the present application, the embodiment provides an art work identification method based on image recognition, the art work identification method based on image recognition comprises the following steps:

[0044] S1, acquiring an art work image and pre-processing the art work image to extract a feature vector of the pre-processed art work image;

[0045] Specifically, acquiring an art work image and pre-processing the art work image means acquiring a true or false art work image from various data sources, the various data sources being digital libraries and museum archives, online art work databases, art trading and auction platforms, social media and artist websites, taking three images from the front, side and back of the art work, color correcting the collected images, adjusting the color temperature and color saturation of the images, using adaptive histogram equalization to enhance the contrast of the images, using a Gaussian filter to remove image noise, adjusting the size of the denoised images to 299*299 pixels, normalizing the image pixel values to the range of 0-1, assigning a unique serial number to the pre-processed art work image and marking the true or false information of the art work and the artist information, storing the pre-processed image dataset into a database.

[0046] Adaptive histogram equalization is a technique for improving image contrast, particularly suitable for enhancing contrast in local regions of an image, which works by dividing the image into many small blocks, applying histogram equalization independently to each small block, adjusting the pixel intensity of each small block so that the histogram distribution of each region is uniform, thereby enhancing the contrast of local regions, applying interpolation methods at the boundaries between blocks to ensure that the processed image transitions naturally and avoids artificial blocky boundary effects, by collecting multi-angle images of artworks and performing advanced preprocessing, the image quality is significantly improved, using adaptive histogram equalization to enhance image contrast and Gaussian filter to remove noise, the visual clarity and detail performance of the image are optimized, not only improving the accuracy of subsequent feature extraction, but also providing high-quality input data for the entire identification system, through color correction and size standardization of the image, ensuring that artwork images of different sources have consistent quality and format before inputting into the model, enhancing the applicability and robustness of the system.

[0047] Further, the feature vector of the pre-processed artwork image is extracted by using the Inception-v3 model to extract the feature vector of the artwork image, loading the pre-trained Inception-v3 model from the deep learning framework library, avoiding loading the top network layer of the model used for classification when loading the model, using the second last convolutional layer in the model as the output layer, using the pre-processed true and false artwork images as training data, enhancing the training data by rotating, scaling and flipping, inputting the training data into the Inception-v3 model, defining the loss function and Adam optimizer for model parameter iterative training until the loss is minimized and stopping iteration to obtain the trained Inception-v3 model, inputting the pre-processed artwork image into the Inception-v3 model to extract the artwork image feature vector B.

[0048] Through the extraction of the pre-processed artwork image feature vector, the accuracy and efficiency of artwork identification are significantly improved, the Inception-v3 model is used to extract the image feature vector, which makes it possible to accurately capture key visual features from artworks, thereby supporting accurate identification of artwork style, age and author, through data augmentation and efficient model training methods, not only the speed of processing large amounts of image data is improved, but also the adaptability of the model to novel artistic styles is enhanced, the robustness and flexibility of the system are improved, making this method not only suitable for basic authenticity identification, but also supporting more complex decision-making, such as style classification and market value assessment, greatly enhancing the decision-making efficiency and information transparency of artwork market participants, promoting the overall development of the artwork market.

[0049] S2, performing reverse image search and calculating the similarity of the user-uploaded artwork image and the artwork image in the database to perform image comparison, constructing an artwork identification model to identify the authenticity of the artwork;

[0050] Specifically, performing reverse image search and calculating the similarity of the user-uploaded artwork image and the artwork image in the database to perform image comparison refers to normalizing the extracted feature vector and inputting the feature vector into a local sensitive hashing algorithm to construct an index. The generated hash code and the corresponding artwork identifier are stored in the index database. Local sensitive hashing is an algorithm for fast data retrieval, especially suitable for processing large-scale data sets and high-dimensional data spaces such as image or text feature vectors. The working principle is that similar objects should have a higher probability of being hashed into the same "bucket", while dissimilar objects have a lower probability of being hashed into the same bucket.

[0051] After receiving the user-uploaded artwork image and preprocessing the artwork image, the artwork image is input into the trained Inception-v3 model to extract the artwork image feature vector A. The index constructed using the local sensitive hashing algorithm is used to quickly search for similar artwork in the database, and the feature similarity of the user-uploaded artwork image and the artwork image in the database is calculated:

[0052]

[0053] Where sim(A, B) is the similarity between the user-uploaded artwork image feature vector A and the artwork image feature vector B in the database, I(A) is the interest vector of the user-uploaded artwork image feature vector A, I(B) is the interest vector of the artwork image feature vector B in the database, and ||I(A)|| and ||I(B)|| are the L2 norms of the interest vectors of the user-uploaded artwork image feature vector A and the artwork image feature vector B in the database, respectively.

[0054] The artwork images are sorted in descending order according to the similarity score, the artwork with the highest similarity is selected, and a list of similar artwork is displayed according to the similarity score, including the image of each artwork, artist information, and similarity score.

[0055] The extracted feature vector A of the user-uploaded artwork image and the feature vector B of the artwork image in the database are combined into a feature vector F as follows:

[0056] F = [A, B].

[0057] The similar art works in the database are quickly retrieved through the reverse image search combined with the local sensitive hashing algorithm, efficient image comparison is realized, the application of the technology makes the identification process more rapid, and the most similar item to the art work to be identified can be found from the huge database in a short time, the similarity score calculated further assists the identification personnel to make accurate judgment, which not only improves the user experience, but also guarantees the fairness and transparency of the identification result, and can greatly reduce the possibility of misjudgment, and provides a fast and reliable identification means for the art work market.

[0058] Further, the art work identification model comprises,

[0059] An art work relation graph is constructed, nodes in the graph structure are defined, each node represents an art work, and the node feature is a feature vector F, edges in the graph structure are defined, and the edges represent the similarity relationship between the art works, and a graph G containing all art works and their similarity relationships is constructed;

[0060] A graph convolution operation is performed on the art work relation graph, and the initial feature vector H 0 of each node is initialized as the node feature F, and multi-layer graph convolution is performed on the feature of each node i to update the feature of each node i:

[0061]

[0062] In the formula, is the feature representation of node i in the l+1 layer, σ is a ReLU activation function, N(i) is a neighbor node set of node i, d i and d j are the number of neighbor nodes of node i and node j respectively, W (l) is a weight matrix of the l layer, is the feature representation of neighbor node j in the l layer, b (l) is a bias term of the l layer;

[0063] After multi-layer graph convolution, the final feature representation M of each node is obtained, the final feature representation M of all nodes is collected, and the feature representation M is input into a fully connected layer of a deep neural network to further extract a high-level feature representation, to obtain a feature representation M' processed by the deep neural network;

[0064] The number of nodes of the input layer of the deep neural network is equal to the dimension of the input feature vector, each fully connected layer of the deep neural network receives the output of the previous layer, further extracts features through linear transformation and nonlinear activation, and adds a ReLU activation function after each fully connected layer to perform nonlinear activation on the output, high-level features are extracted and fused layer by layer through the multi-layer fully connected network, and the output layer of the deep neural network is defined to output the final high-level feature representation.

[0065] An artistic work identification model is built based on a support vector machine:

[0066] D = sign(ω * M' + b),

[0067] where D is the model output, M' is the feature representation after processing by the deep neural network, ω is the model weight, b is the bias term, sign is the classification function, +1 represents real, and -1 represents fake;

[0068] The image data set stored in the database is used as the model training set, and the model training set is used for model training, and the model optimization objective is defined as:

[0069]

[0070] where Dq is the output of the qth training data of the model, M'q is the qth training data, subject is the definition of optimization constraint condition, represents that the constraint condition is established for all q;

[0071] After solving the optimization objective to obtain the model parameters, the artistic work identification model is obtained by substituting the model.

[0072] By constructing an artistic work relationship graph, the nodes and edges in the graph are defined to represent the artistic works and their similarity relationships. The graph convolution network (GCN) is used to update and deepen the feature vectors of the artistic works at multiple levels, achieving high abstraction and refinement of the features of the artistic works. This not only enhances the performance of the model in processing complex artistic data, but also improves the ability of the identification model to distinguish between real and fake artistic works. By further extracting features through a deep neural network and combining a support vector machine for final classification, the model can effectively distinguish between real and fake artistic works. By applying graph convolution to update the features of the artistic works, the model can more comprehensively understand the subtle differences between artistic works, thereby achieving higher identification accuracy.

[0073] Furthermore, for artistic work authenticity identification, the feature vector of the user-uploaded artistic work image is combined with the feature vector of the artistic work image with the highest similarity in the database. After graph convolution operation and deep neural network processing, the final feature representation M' is obtained. The final feature representation M' is input into the artistic work identification model to obtain the artistic work identification result.

[0074] By comparing the feature vectors of user-uploaded artworks with the most similar feature vectors in the database, and using the previously constructed discrimination model to output the authenticity results of the artworks, this method allows for quick and accurate identification of the authenticity status of artworks, particularly in distinguishing between art replicas and original works. Through this comparison and identification process, the history and authenticity of artworks can be effectively revealed, which is of great significance for the art market and cultural heritage protection.

[0075] S3, labeling the artworks after authenticity discrimination, securely storing the data and implementing access control;

[0076] Specifically, labeling the artworks after authenticity discrimination means labeling the artworks according to the discrimination results,

[0077] If the discriminated artwork is a real artwork, the real artwork is labeled as a "real product" tag, and a digital certificate is generated for each artwork identified as a real product, including the artist information of the artwork, the historical ownership of the artwork, and the sale and exhibition history;

[0078] If the discriminated artwork is a fake artwork, the fake artwork is labeled as a "fake product" tag, and the record includes the identification date and confidence level, and provides the reasons for identifying as fake, including the use of non-original materials and the obvious inconsistency in style with the original.

[0079] According to the discrimination results, real artworks are labeled as "real products" and generate digital certificates including artist information, historical ownership, exhibition history, etc., which increases the information transparency and traceability of artworks, benefiting both buyers and sellers. For artworks identified as fake, by labeling them as "fake products" and recording specific reasons for the forgery (such as the use of non-original materials or inconsistent style), key information is provided, which helps to regulate the art market and improve the effectiveness of art identification.

[0080] Further, securely storing the data and implementing access control means storing the information of the artworks, the discrimination results, and the timestamps in the database, and setting up security access measures. The database will store the data in the cloud backup, and periodically detect the integrity of the stored data and the backup data. After the detection is completed, an integrity detection record is generated and stored in the database, allowing authorized users to query the art identification results through the user interface.

[0081] By storing all the authentication data in a secure database and implementing access control to ensure that only authorized users can access sensitive information, and by implementing cloud backup and integrity detection to enhance the security and persistence of data, the integrity and confidentiality of the artwork data are maintained, and strong protection is provided against unauthorized access and misuse of the artwork data.

[0082] Embodiment 2

[0083] For the second embodiment of the present application, the embodiment is different from the previous embodiment in that:

[0084] The functions described above can be implemented in hardware or software stored on computer-readable storage medium. Based on such an understanding, the technical solutions of the present application essentially or the part that contributes to the prior art or the part of the technical solutions can be embodied in the form of a software product. The computer software product is stored in a storage medium, and includes a number of instructions for causing a computer device (which can be a personal computer, a server, or a network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of the present application. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), magnetic disk or optical disk, and various program code storage media.

[0085] The logic and / or steps represented in the flowchart or otherwise described herein, for example, can be considered a list of executable instructions for implementing logic functions, and can be specifically embodied in any computer-readable medium for use by an instruction execution system, apparatus or device, such as a computer-based system, a system including a processor, or other system that can fetch instructions from an instruction execution system, apparatus or device and execute the instructions, or in conjunction with these instruction execution systems, apparatus or devices. For the purposes of this specification, a "computer-readable medium" can be any device that can contain, store, communicate, propagate or transport programs for use by an instruction execution system, apparatus or device, or in conjunction with these instruction execution systems, apparatus or devices.

[0086] More specific examples (a non-exhaustive list) of the computer-readable medium include the following: an electrical connection (electronic) having one or more wires, a portable computer diskette (magnetic), a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or Flash memory), an optical fiber device, and a portable compact disc read-only memory (CDROM). Additionally, the computer-readable medium can be paper or another suitable medium upon which the program is printed, as the program can be electronically captured, for example, via an optical scanner, then compiled, interpreted, or otherwise processed, and stored in a computer memory in a form that is then employable by a computer. Examples of computer-readable media include but are not limited to portable computer disks, hard disks, RAM, ROM, EEPROM, and optical

[0087] It should be understood that aspects of the application can be implemented in hardware, software, firmware or combinations thereof. In the embodiments described above, various steps or methods can be implemented, in part, or in whole, in software and / or firmware that is stored in memory and executed by a suitable instruction execution system. For example, if implemented in hardware, and in another embodiment, any of the following technology, known in the art, can be employed for implementation: discrete logic circuitry having logic gates for implementing logic functions upon data signals, application specific integrated circuits having logic gates, field programmable gate arrays (FPGAs), and / or programmable logic arrays (PLAs), etc.

Claims

1. An art identification method based on image recognition, characterized in that: include, Acquire the image of the artwork and preprocess it, then extract the feature vector B of the preprocessed artwork image; Perform reverse image search and calculate the similarity between user-uploaded artwork images and artwork images in the database for image comparison. This includes preprocessing the user-uploaded artwork images and then inputting the artwork images into the trained Inception-v3 model to extract the artwork image feature vector A. The feature vector A extracted from the user-uploaded artwork image and the feature vector B from the artwork image in the database are combined into a feature vector F: F = [A, B]; Constructing an art authentication model to identify the authenticity of artworks; After verifying the authenticity of the artworks, the data is labeled, securely stored, and access is controlled. The constructed artwork identification model includes, Construct an art work relationship graph, define nodes in the graph structure, each node represents an art work, node features are feature vectors F, define edges in the graph structure, edges represent similarity relationships between art works, and construct a graph G containing all art works and their similarity relationships; Perform graph convolution on the relationship graph of the artworks, and convert the initial feature vector H of each node into a graph. 0 Initialize the features to node F, perform multi-layer graph convolution, and update the features of each node i using graph convolution: In the formula, σ is the feature representation of node i in layer l+1, σ is the ReLU activation function, N(i) is the set of neighboring nodes of node i, and d i and d j These are the number of neighbors of node i and node j, respectively, W. (l) It is the weight matrix of the l-th layer. b is the feature representation of neighbor node j in layer l. (l) It is the bias term of the l-th layer; After multi-layer graph convolution, the final feature representation M of each node is obtained. The final feature representation M of all nodes is collected and input into the fully connected layer of the deep neural network to further extract high-level feature representations, resulting in the feature representation M′ after processing by the deep neural network. Building an art identification model based on support vector machines: D = sign(ω*M′+b), Where D is the model output, M′ is the feature representation after processing by the deep neural network, ω is the model weight, b is the bias term, sign is the classification function, +1 represents real and -1 represents fake; The image dataset stored in the database is used as the model training set. The model is trained using the training set, and the model optimization objective is defined as follows: Where Dq is the output of the model on the q-th training data, M'q is the output of the q-th training data, and subject defines the optimization constraints. This means that all q constraints are true; The model parameters are obtained by solving the optimization objective, and then input into the model to obtain the artwork identification model.

2. The art identification method based on image recognition as described in claim 1, characterized in that: The process of acquiring and preprocessing artwork images involves obtaining genuine and counterfeit artwork images from various data sources, taking three images each from the front, side, and back of the artwork, performing color correction on the acquired images, adjusting the color temperature and saturation, using adaptive histogram equalization to enhance image contrast, using a Gaussian filter to remove image noise, adjusting the size of the denoised image to 299*299 pixels, normalizing the image pixel values ​​to the range of 0-1, assigning a unique serial number to the preprocessed artwork images to identify their authenticity and artist information, and storing the preprocessed image dataset in a database.

3. The artwork identification method based on image recognition as described in claim 2, characterized in that: The extraction of the preprocessed artwork image feature vector refers to using the Inception-v3 model to extract the feature vector of the artwork image. This involves loading a pre-trained Inception-v3 model from a deep learning framework library, using the penultimate convolutional layer as the output layer, and using preprocessed real and fake artwork images as training data. The training data is then enhanced through rotation, scaling, and flipping. The training data is input into the Inception-v3 model, and a loss function and Adam optimizer are defined for iterative training of the model parameters until the loss is minimized. The iteration is then stopped to obtain the trained Inception-v3 model. Finally, the preprocessed artwork image is input into the Inception-v3 model to extract the artwork image feature vector B.

4. The artwork identification method based on image recognition as described in claim 3, characterized in that: The process of performing reverse image search and calculating the similarity between user-uploaded artwork images and artwork images in the database for image comparison involves standardizing the extracted feature vectors, inputting the feature vectors into a local sensitive hash algorithm to build an index, and storing the generated hash codes and corresponding artwork identifiers in the index database. The system receives user-uploaded artwork images, preprocesses them, and then inputs them into a trained Inception-v3 model to extract feature vectors A. A locality-sensitive hashing (LSH) algorithm is used to quickly search for similar artworks in the database, and the feature similarity between the user-uploaded artwork image and the artwork images in the database is calculated. Where sim(A, B) is the similarity between the feature vector A of the artwork image uploaded by the user and the feature vector B of the artwork image in the database, I(A) is the interest vector of the feature vector A of the artwork image uploaded by the user, I(B) is the interest vector of the feature vector B of the artwork image in the database, and ||I(A)|| and ||I(B)|| are the L2 norms of the interest vectors of the feature vectors A of the artwork image uploaded by the user and the feature vector B of the artwork image in the database, respectively. The artwork images are sorted in descending order of similarity score. The artwork with the highest similarity score is selected and a list of similar artworks is displayed based on the similarity score. This list includes the image of each artwork, artist information, and similarity score. The feature vector A extracted from the user-uploaded artwork image and the feature vector B from the artwork image in the database are combined into a feature vector F: F = [A, B].

5. The artwork identification method based on image recognition as described in claim 4, characterized in that: The process of identifying the authenticity of artworks involves combining the feature vectors of user-uploaded artwork images with the feature vectors of the most similar artwork images in the database. After processing through graph convolution operations and deep neural networks, the final feature representation M′ is obtained. The final feature representation M′ is then input into the artwork identification model to obtain the artwork identification result.

6. The method for identifying artworks based on image recognition as described in claim 5, characterized in that: The annotation of artworks after authentication refers to annotating the artworks based on the authentication results. If an artwork is identified as a genuine artwork, it will be labeled as "authentic" and a digital certificate will be generated for each artwork identified as authentic, including the artist's information, the artwork's historical ownership, and its sale and exhibition history. If an artwork is identified as a forgery, it will be labeled "forgery," and the identification date, confidence level, and reasons for the identification will be provided.

7. The artwork identification method based on image recognition as described in claim 6, characterized in that: The process of securely storing data and implementing access control involves storing the artwork's information, authentication results, and timestamp in a database, setting up secure access measures, backing up the stored data to the cloud, and periodically performing integrity checks on the stored and backup data. After the checks are completed, integrity check records are generated and synchronously stored in the database, allowing authorized users to query the artwork's authentication results through the user interface.

8. A computer device, comprising: Memory and processor; The memory stores a computer program, characterized in that: when the processor executes the computer program, it implements the steps of the image recognition-based artwork identification method according to any one of claims 1 to 7.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by the processor, it implements the steps of the image recognition-based artwork identification method according to any one of claims 1 to 7.

Images