Multi-modal artistic style comparison and identification method

Abstract

The invention relates to the technical field of artistic style identification, in particular to a multi-modal artistic style comparison identification method, which comprises the following steps of: intelligently matching an artistic style feature acquisition scheme by acquiring style category text feature data of a target artistic work and combining preset different style work feature acquisition schemes; collecting target artistic style feature data according to the artistic style feature acquisition scheme; clustering processing is carried out on the collected artistic style feature data set, an artistic style clustering feature data matrix is generated, corresponding artistic style type text feature data is set, then an artistic style analysis model is constructed, the artistic style of the target artistic work is analyzed through the artistic style analysis model, and the artistic style of the target artistic work is analyzed. And generating target artistic style type text feature data, and accurately searching artistic style comparison data of the target artistic works from the artistic style clustering feature data matrix in combination with an intelligent optimization algorithm to realize comparison and identification of the multi-modal artistic styles.

Description

Technical Field

[0001] The present invention relates to the technical field of artistic style identification, and in particular to a multimodal artistic style comparison and identification method. Background Art

[0002] With the continuous development of Internet technology and the increasing diversification of art forms, traditional art style identification technology can no longer meet people's needs, and it has become increasingly important to identify different art styles more accurately and efficiently.

[0003] Existing art style identification technologies often rely on single-modal data and are unable to dynamically adjust multi-modal feature data acquisition schemes according to artistic genres, making it difficult to comprehensively and accurately analyze the stylistic characteristics of artworks. At the same time, art style analysis often uses feature data, which is highly subjective, resulting in poor accuracy and stability in the art style identification results. Summary of the Invention

[0004] In response to the problems in the related art, the present invention provides a multimodal art style comparison and identification method to overcome the above-mentioned technical problems existing in the existing related art.

[0005] To solve the above technical problems, the present invention provides the following technical solution: a multimodal art style comparison and identification method, comprising the following steps:

[0006] S1. Collecting text feature data of the genre category of the target artwork;

[0007] S2. Matching the genre category text feature data with feature acquisition schemes for works of different genres to generate an artistic style feature acquisition scheme for the target artwork;

[0008] S3. Collecting artistic style feature data of a target artwork according to the artistic style feature acquisition scheme to obtain target artistic style feature data;

[0009] S4, collecting artistic style feature data of different artworks corresponding to the genre category text feature data to obtain an artistic style feature data set, performing clustering processing on the artistic style feature data set to generate an artistic style cluster feature data matrix, and setting corresponding artistic style type text feature data;

[0010] S5. Constructing an artistic style analysis model based on the artistic style cluster feature data matrix and the artistic style type text feature data;

[0011] S6. Analyze the artistic style of the target artwork according to the artistic style analysis model and the target artistic style feature data to generate target artistic style type text feature data;

[0012] S7: Match the target artistic style feature data with the artistic style cluster feature data corresponding to the target artistic style type text feature data to generate artistic style comparison data of the target artwork.

[0013] Preferably, the specific steps of collecting the genre category text feature data of the target artwork are as follows:

[0014] S11. Collect genre category text feature data A of the target artwork online through an art style identification platform.

[0015] Preferably, the specific steps of matching the genre category text feature data with feature acquisition schemes for works of different genres to generate an artistic style feature acquisition scheme for a target artwork are as follows:

[0016] S21. Establish a set of solutions for acquiring features of works of different genres in, represents the artistic style feature acquisition scheme corresponding to the i-th artwork genre, and k represents the total number of artwork genres;

[0017] The genres of the artistic works include but are not limited to painting, sculpture, music and film and television;

[0018] The artistic style feature acquisition scheme represents the multimodal artistic style feature data required to be acquired for different genres of artworks;

[0019] The multimodal art style feature data includes but is not limited to any one or more of image feature data, audio feature data, three-dimensional feature data, text feature data, and video feature data;

[0020] S22, performing character feature matching on the genre category text feature data A and the different genre work feature acquisition schemes in the different genre work feature acquisition scheme set using the Sunday matching algorithm;

[0021] If the genre text feature data A successfully matches any one of the feature acquisition schemes for different genre works in the feature acquisition scheme set for different genre works, the character feature matching is terminated, and the feature acquisition scheme for different genre works that matches the genre text feature data A is data identified to generate the artistic style feature acquisition scheme A for the target artwork. mubiao ;

[0022] Otherwise, continue to perform character adjustment matching until the genre category text feature data A is successfully matched with any one of the feature acquisition schemes for different genre works in the feature acquisition scheme set for different genre works.

[0023] By collecting the genre category text feature data of the target artwork and combining it with the Sunday matching algorithm, the artistic style feature acquisition scheme suitable for the target artwork is accurately matched from the feature acquisition schemes of works of different genres, avoiding the blind collection of all artistic style feature data, and determining the type of data to be collected in a targeted manner according to the genre of the work, avoiding additional resource consumption, and at the same time improving the efficiency of artistic style comparison and identification.

[0024] Preferably, the specific steps of collecting the artistic style feature data of the target artwork according to the artistic style feature acquisition scheme to obtain the target artistic style feature data are as follows:

[0025] S31, obtaining solution A based on the artistic style characteristics mubiao Different sensors are selected to collect multimodal artistic style feature data of the target artwork, and a target artistic style feature dataset B = {b1, b2, ..., b i ,…,b l}, where b i represents the i-th multimodal art style feature data of the target artwork, and l represents the total number of multimodal art style feature data of the target artwork.

[0026] Preferably, the specific steps of collecting artistic style feature data of different artworks corresponding to the genre category text feature data to obtain an artistic style feature data set, performing clustering processing on the artistic style feature data set to generate an artistic style cluster feature data matrix, and setting the corresponding artistic style type text feature data are as follows:

[0027] S41, collect multimodal art style feature data of different works of art corresponding to the genre category text feature data online through the art style identification platform, and generate an art style feature dataset C = {c1, c2, ..., c i ,…,c p}, where c i represents the artistic style feature data of the i-th artwork collected by the artistic style identification platform, and p represents the total number of artworks collected by the multimodal artistic style identification platform;

[0028] S42, using an agglomerative hierarchical clustering algorithm to perform data clustering processing on the artistic style feature data in the artistic style feature data set to generate an artistic style cluster feature data matrix as follows:

[0029]

[0030] in, represents the multimodal art style feature data of the jth artwork contained in the i-th art style collected through the multimodal art style identification platform, d i represents the total number of artworks contained in the i-th category of art style collected through the multimodal art style identification platform; the data clustering process is as follows:

[0031] S421: Treat each artistic style feature data in the artistic style feature data set as a cluster, calculate the distance between each cluster, and obtain a distance matrix. The distance calculation formula is as follows:

[0032]

[0033] Where D(x,y) represents the distance between vector x and vector y, x and y represent the l-dimensional feature vectors of any two cluster centers, x i Represents the value of vector x in the i-th dimension, y i Represents the value of vector y in the i-th dimension;

[0034] S422, merging the two clusters corresponding to the minimum values in the distance matrix to obtain a new cluster, calculating the distance between the new cluster and other clusters, and updating the distance matrix according to the calculation result;

[0035] S423, setting a distance threshold;

[0036] If all distances in the distance matrix are greater than the distance threshold, the data clustering process is terminated to obtain the artistic style clustering feature data matrix.

[0037] Otherwise, return to S422 until all distances in the distance matrix are greater than the distance threshold;

[0038] S43, annotating the corresponding art style category text feature data for each art style in the multimodal art style clustering feature data matrix, and generating an art style category text feature dataset E = {e1, e2, ..., e i ,…,e o}, where e i Represents the text feature data corresponding to the i-th artistic style.

[0039] By processing the collected art style feature data set with an agglomerative hierarchical clustering algorithm, an art style cluster feature data matrix is generated, which can intuitively identify the differences between the acquired data, achieve scientific classification of art styles, and provide a data basis for subsequent steps.

[0040] Preferably, the specific steps of constructing an artistic style analysis model based on the artistic style cluster feature data matrix and the artistic style type text feature data are as follows:

[0041] S51, constructing an initial SVM model, setting a training data ratio, and performing data partitioning on the artistic style cluster feature data matrix and the artistic style type text feature data set according to the training data ratio to obtain an artistic style cluster feature training data matrix, an artistic style type text feature training data set, an artistic style cluster feature test data matrix, and an artistic style type text feature test data set;

[0042] S52, setting a training error threshold and a maximum number of training times, inputting the artistic style cluster feature training data matrix and the artistic style type text feature training data set as training data and training label data into the initial SVM model for training, adjusting initial parameters of the initial SVM model according to the training results, and then continuing to train the SVM with the adjusted parameters until the training error is less than the training error threshold or the number of training times is greater than the maximum number of training times, thereby obtaining a trained SVM model;

[0043] S53. Set a test accuracy threshold, input the artistic style cluster feature test data matrix and the artistic style type text feature test data set as test data and test label data respectively into the trained SVM model for testing, calculate the test accuracy, and if the test accuracy is greater than the test accuracy threshold, obtain the artistic style analysis model; otherwise, return to S52 and retrain until the test accuracy is greater than the test accuracy threshold.

[0044] An art style analysis model is constructed based on the art style clustering feature data matrix and the corresponding art style type text feature data, which provides a good tool for analyzing the style of artworks and improves the reliability and accuracy of art style.

[0045] Preferably, the specific steps of analyzing the artistic style of the target artwork based on the artistic style analysis model and the target artistic style feature data to generate target artistic style type text feature data are as follows:

[0046] S61: Input the target artistic style feature data into the artistic style analysis model to analyze the artistic style of the target artwork, and generate target artistic style type text feature data F. mubiao .

[0047] Preferably, the specific steps of matching the target artistic style feature data with the artistic style cluster feature data corresponding to the target artistic style type text feature data to generate artistic style comparison data of the target artwork are as follows:

[0048] S71: Performing artistic style feature matching processing on the target artistic style feature data and the artistic style cluster feature data corresponding to the target artistic style type text feature data in the artistic style cluster feature data by using an intelligent optimization algorithm, searching for artistic style cluster feature data that matches the target artistic style feature data and performing data identification, thereby generating artistic style comparison data Q of the target artwork. mubiao ;

[0049] S711. Construct an artistic style feature to identify the water flow population, set the population size to N, the current number of iterations to t, and the maximum number of iterations to t. max The dimension of the search space for the art style clustering feature data is P;

[0050] Using the artistic style cluster feature data corresponding to the target artistic style type text feature data in the artistic style cluster feature data as an artistic style cluster feature data search space, randomly generating N artistic style cluster feature data in the artistic style cluster feature data search space, each artistic style cluster feature data corresponding to an artistic style feature identification water flow individual in the artistic style feature identification water flow population;

[0051] S712. Calculate the fitness value between each individual artistic style feature recognition water flow in the artistic style feature recognition water flow population and the target artistic style feature data, sort each individual artistic style feature recognition water flow in the artistic style feature recognition water flow population from largest to smallest according to the fitness value, and select the artistic style feature recognition water flow individual with the highest fitness value as the current optimal individual; the fitness value calculation formula is as follows:

[0052]

[0053] Among them, f i represents the fitness value between the i-th artistic style feature recognition water flow individual in the artistic style feature recognition water flow population and the target artistic style feature data, m i represents the characteristic vector of the artistic style cluster feature data corresponding to the i-th artistic style feature identification water flow individual in the artistic style feature identification water flow population, n represents the characteristic vector of the target artistic style feature data, and φ represents a correction value;

[0054] S713, calculating the permeability coefficient α of each individual water flow identified by the artistic style feature in the water flow population identified by the artistic style feature in the search space of the artistic style cluster feature data i ; The calculation formula of permeability coefficient is as follows:

[0055]

[0056] Among them, α i represents the permeability coefficient of the i-th individual water flow identified by the artistic style feature in the water flow population, f max and f min Respectively represent the fitness values corresponding to the best position and the worst position in the water flow population identified by the artistic style feature;

[0057] Each individual water flow identified by the artistic style feature in the water flow population identified by the artistic style feature controls its own behavior and updates its position in the artistic style cluster feature data search space according to the permeability coefficient. The position update formula is as follows:

[0058]

[0059] in, represents the position of the i-th individual in the population of artistic style feature recognition water flow after the position is updated, M i and M j Respectively represent the current positions of the i-th and j-th art style feature identification water flow individuals in the art style feature identification water flow population, r1, r2 and r all represent random numbers uniformly distributed between (0,1), λ and η respectively represent the upper and lower limits of the art style clustering feature data search space, M best represents the position of the current optimal individual, β represents the water consumption coefficient that changes with the number of current iterations, ω t represents the adaptive step size factor, and Among them, ω max and ω min Respectively represent the maximum and minimum adaptive step sizes;

[0060] S714: randomly select an artistic style feature identification water flow individual from the artistic style feature identification water flow population according to probability, and update the position of the selected artistic style feature identification water flow individual in the artistic style cluster feature data search space according to the position of the current optimal individual; the position update formula is as follows:

[0061]

[0062] in, represents the position of the oth artistic style feature identification water flow individual randomly selected from the artistic style feature identification water flow population according to probability after the position update, and r3 and r4 both represent random numbers uniformly distributed between (0,1);

[0063] S715, calculating the fitness value of each individual artistic style feature identification water flow in the artistic style feature identification water flow population after the position is updated, if the fitness value of the individual artistic style feature identification water flow after the position is updated is greater than the original fitness value, replacing the original position with the new position; otherwise, retaining the original position;

[0064] Re-sorting the individual artistic style feature identification water flow individuals in the artistic style feature identification water flow population according to their fitness values from large to small, and selecting the artistic style feature identification water flow individual with the highest fitness value as the new current optimal individual;

[0065] S716: Determine whether the current number of iterations t is greater than or equal to the maximum number of iterations t max , if the current number of iterations t is greater than or equal to the maximum number of iterations t max , the current optimal individual is taken as the global optimal solution; otherwise, the current iteration number t is increased by 1, and the process returns to S713;

[0066] S72: Combine the target artistic style type text feature data, the target artistic style feature data set, and the artistic style comparison data to generate artistic style comparison identification data W={F mubiao ,C,Q mubiao}, pushing the art style comparison and identification data to the art style identification platform through a wireless communication network and displaying it on a display screen.

[0067] The artistic style of the target artwork is accurately analyzed through the artistic style analysis model, and the matching range in the artistic style clustering feature data matrix is narrowed. Matching is performed based on the narrowed matching range combined with the flood optimization algorithm. Through multiple iterations, the multimodal artistic style feature data that best matches the target artwork is accurately searched, and the comparison of multimodal artistic styles is achieved, which improves the speed and efficiency of the matching process. At the same time, an adaptive step size factor is set to control the search range during the execution of the algorithm, so that the search range is larger in the early stage of the iterative process and smaller in the later stage, thereby improving the search efficiency of the algorithm.

[0068] The present invention also includes a multimodal art style comparison and identification system, comprising a genre category text feature data acquisition module, an art style feature acquisition scheme matching module, a target art style feature data acquisition module, a target art style feature data clustering module, an art style analysis model construction module, a target artwork style analysis module, and an art style comparison data matching module;

[0069] The genre category text feature data acquisition module collects genre category text feature data of the target artwork online through the art style identification platform;

[0070] The artistic style feature acquisition scheme matching module performs character feature matching on the genre category text feature data and the feature acquisition schemes for works of different genres using a Sunday matching algorithm to generate an artistic style feature acquisition scheme for a target artwork;

[0071] The target art style feature data acquisition module selects different sensors according to the art style feature acquisition scheme to collect multimodal art style feature data of the target artwork to obtain target art style feature data;

[0072] The target art style feature data clustering module collects multimodal art style feature data of different artworks corresponding to the genre category text feature data online through the art style identification platform to generate an art style feature data set, and clusters the art style feature data set using an agglomerative hierarchical clustering algorithm to generate an art style cluster feature data matrix, and sets corresponding art style type text feature data;

[0073] The art style analysis model construction module trains and tests the initial SVM model based on the art style cluster feature data matrix and the art style type text feature data to obtain an art style analysis model;

[0074] The target artwork style analysis module inputs the target artwork style feature data into the art style analysis model to analyze the art style of the target artwork and generate target art style type text feature data;

[0075] The artistic style comparison data matching module matches the target artistic style feature data with the artistic style clustering feature data corresponding to the target artistic style type text feature data through an intelligent optimization algorithm, generates artistic style comparison data of the target artwork, and constructs artistic style comparison identification data, which is pushed to the artistic style identification platform through a wireless communication network and displayed on a display screen.

[0076] By means of the above technical solution, the present invention provides a multimodal art style comparison and identification method, which has at least the following beneficial effects:

[0077] 1. The present invention obtains the genre category text feature data of the target artwork, and intelligently matches the artistic style feature acquisition scheme in combination with preset feature acquisition schemes of different genre works, and collects the target artistic style feature data according to the artistic style feature acquisition scheme; clusters the collected artistic style feature data set to generate an artistic style cluster feature data matrix, and sets corresponding artistic style type text feature data, and then constructs an artistic style analysis model, analyzes the artistic style of the target artwork through the artistic style analysis model, generates target artistic style type text feature data, and combines with an intelligent optimization algorithm to accurately match the artistic style comparison data of the target artwork from the artistic style cluster feature data matrix, thereby realizing comparative identification of multimodal artistic styles.

[0078] 2. The present invention collects the genre category text feature data of the target artwork and combines it with the Sunday matching algorithm to accurately match the artistic style feature acquisition scheme suitable for the target artwork from the feature acquisition schemes of works of different genres, thereby avoiding the blind collection of all artistic style feature data, and determining the type of data to be collected in a targeted manner according to the genre of the work, avoiding additional resource consumption, and at the same time improving the efficiency of artistic style comparison and identification.

[0079] 3. The present invention processes the collected artistic style feature data set using an agglomerative hierarchical clustering algorithm to generate an artistic style clustering feature data matrix, intuitively identifies the differences between the acquired data, realizes the scientific classification of artistic styles, and provides a data basis for subsequent steps; and sets the corresponding artistic style type text feature data, on this basis constructs an artistic style analysis model, which provides a good tool for analyzing the style of artworks and improves the reliability and accuracy of artistic styles.

[0080] 4. The present invention accurately analyzes the artistic style of the target artwork through the artistic style analysis model, narrows the matching range in the artistic style clustering feature data matrix, and performs matching based on the narrowed matching range combined with the flood optimization algorithm. Through multiple iterations, the multimodal artistic style feature data that best matches the target artwork is accurately searched, thereby realizing the comparison of multimodal artistic styles and improving the speed and efficiency of the matching process. At the same time, an adaptive step size factor is set to control the search range during the execution of the algorithm, so that the search range in the early stage of the iterative process is larger and the search range in the later stage is smaller, thereby improving the search efficiency of the algorithm. BRIEF DESCRIPTION OF THE DRAWINGS

[0081] In order to more clearly illustrate the technical solutions of the embodiments of the invention, the following briefly introduces the drawings required for describing the embodiments. Obviously, the drawings described below are only some embodiments of the invention. For ordinary technicians in this field, they can also obtain drawings based on these drawings without paying any creative work.

[0082] Figure 1 A flowchart of the multimodal art style comparison and identification method provided by the present invention;

[0083] Figure 2 This is a module diagram of the multimodal art style comparison and identification system provided by the present invention. DETAILED DESCRIPTION

[0084] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention.

[0085] Example 1 is as follows:

[0086] In view of the problem that the existing art style identification technology often combines past experience with the single modal data of the artwork to conduct art style analysis, resulting in poor accuracy and stability of the art style identification results. This embodiment proposes a multimodal art style comparison and identification method, which can dynamically adjust the data acquisition scheme based on the genre category text feature data of the target art work in combination with the intelligent search algorithm. Based on the data acquisition scheme, the target art style feature data is obtained, and an art style analysis model is constructed on the basis of the art style clustering feature data matrix and the art style type text feature data. The art style of the target art work is analyzed, and the target art style type text feature data is generated. In combination with the intelligent optimization algorithm, the art style comparison data of the target art work is accurately matched from the art style clustering feature data matrix, thereby realizing multimodal art style comparison and identification. Figure 1 As shown, the method includes the following steps:

[0087] S1. Collecting text feature data of the genre category of the target artwork. S11. Collecting text feature data A of the genre category of the target artwork online through an art style identification platform.

[0088] S2. Matching the genre category text feature data with feature acquisition schemes for works of different genres to generate an artistic style feature acquisition scheme for the target artwork. As a specific implementation plan of this method, the detailed scheme of this step is as follows:

[0089] S21. Establish a set of solutions for acquiring features of works of different genres in, represents the artistic style feature acquisition scheme corresponding to the i-th artwork genre, and k represents the total number of artwork genres;

[0090] The genres of the artistic works include but are not limited to painting, sculpture, music and film and television;

[0091] The artistic style feature acquisition scheme represents the multimodal artistic style feature data required to be acquired for different genres of artworks;

[0092] The multimodal art style feature data includes but is not limited to any one or more of image feature data, audio feature data, three-dimensional feature data, text feature data, and video feature data;

[0093] S22, performing character feature matching on the genre category text feature data A and the different genre work feature acquisition schemes in the different genre work feature acquisition scheme set using the Sunday matching algorithm;

[0094] If the genre text feature data A successfully matches any one of the feature acquisition schemes for different genre works in the feature acquisition scheme set for different genre works, the character feature matching is terminated, and the feature acquisition scheme for different genre works that matches the genre text feature data A is data identified to generate the artistic style feature acquisition scheme A for the target artwork. mubiao ;

[0095] Otherwise, continue to perform character adjustment matching until the genre category text feature data A is successfully matched with any one of the feature acquisition schemes for different genre works in the feature acquisition scheme set for different genre works.

[0096] S3. Collect the artistic style feature data of the target artwork according to the artistic style feature acquisition scheme to obtain the target artistic style feature data. S31. Collect the artistic style feature data of the target artwork according to the artistic style feature acquisition scheme A. mubiao Different sensors are selected to collect multimodal artistic style feature data of the target artwork, and a target artistic style feature dataset B = {b1, b2, ..., b i ,…,b l}, where b i represents the i-th multimodal art style feature data of the target artwork, and l represents the total number of multimodal art style feature data of the target artwork.

[0097] S4. Collecting artistic style feature data of different artworks corresponding to the genre category text feature data to obtain an artistic style feature dataset, performing clustering processing on the artistic style feature dataset to generate an artistic style cluster feature data matrix, and setting corresponding artistic style type text feature data. As a specific implementation plan of this method, the detailed plan of this step is as follows:

[0098] S41, collect multimodal art style feature data of different works of art corresponding to the genre category text feature data online through the art style identification platform, and generate an art style feature dataset C = {c1, c2, ..., c i ,…,c p}, where c i represents the artistic style feature data of the i-th artwork collected by the artistic style identification platform, and p represents the total number of artworks collected by the multimodal artistic style identification platform;

[0099] S42, using an agglomerative hierarchical clustering algorithm to perform data clustering processing on the artistic style feature data in the artistic style feature data set to generate an artistic style cluster feature data matrix as follows:

[0100]

[0101] in, represents the multimodal art style feature data of the jth artwork contained in the i-th art style collected through the multimodal art style identification platform, d i represents the total number of artworks contained in the i-th category of art style collected through the multimodal art style identification platform; the data clustering process is as follows:

[0102] S421: Treat each artistic style feature data in the artistic style feature data set as a cluster, calculate the distance between each cluster, and obtain a distance matrix. The distance calculation formula is as follows:

[0103]

[0104] Where D(x,y) represents the distance between vector x and vector y, x and y represent the l-dimensional feature vectors of any two cluster centers, x i Represents the value of vector x in the i-th dimension, y i Represents the value of vector y in the i-th dimension;

[0105] S422, merging the two clusters corresponding to the minimum values in the distance matrix to obtain a new cluster, calculating the distance between the new cluster and other clusters, and updating the distance matrix according to the calculation result;

[0106] S423, setting a distance threshold;

[0107] If all distances in the distance matrix are greater than the distance threshold, the data clustering process is terminated to obtain the artistic style clustering feature data matrix.

[0108] Otherwise, return to S422 until all distances in the distance matrix are greater than the distance threshold;

[0109] S43, annotating the corresponding art style category text feature data for each art style in the multimodal art style clustering feature data matrix, and generating an art style category text feature dataset E = {e1, e2, ..., e i ,…,e o}, where e i Represents the text feature data corresponding to the i-th artistic style.

[0110] S5: Constructing an artistic style analysis model based on the artistic style cluster feature data matrix and the artistic style type text feature data. As a specific implementation plan of this method, the detailed plan of this step is as follows:

[0111] S51, constructing an initial SVM model, setting a training data ratio, and performing data partitioning on the artistic style cluster feature data matrix and the artistic style type text feature data set according to the training data ratio to obtain an artistic style cluster feature training data matrix, an artistic style type text feature training data set, an artistic style cluster feature test data matrix, and an artistic style type text feature test data set;

[0112] S52, setting a training error threshold and a maximum number of training times, inputting the artistic style cluster feature training data matrix and the artistic style type text feature training data set as training data and training label data into the initial SVM model for training, adjusting initial parameters of the initial SVM model according to the training results, and then continuing to train the SVM with the adjusted parameters until the training error is less than the training error threshold or the number of training times is greater than the maximum number of training times, thereby obtaining a trained SVM model;

[0113] S53. Set a test accuracy threshold, input the artistic style cluster feature test data matrix and the artistic style type text feature test data set as test data and test label data respectively into the trained SVM model for testing, calculate the test accuracy, and if the test accuracy is greater than the test accuracy threshold, obtain the artistic style analysis model; otherwise, return to S52 and retrain until the test accuracy is greater than the test accuracy threshold.

[0114] S6. Analyze the artistic style of the target artwork based on the artistic style analysis model and the target artistic style feature data to generate target artistic style type text feature data. S61. Input the target artistic style feature data into the artistic style analysis model to analyze the artistic style of the target artwork to generate target artistic style type text feature data F. mubiao .

[0115] S7, matching the target artistic style feature data with the artistic style cluster feature data corresponding to the target artistic style type text feature data to generate artistic style comparison data of the target artwork. As a specific implementation plan of this method, the detailed plan of this step is as follows:

[0116] S71: Performing artistic style feature matching processing on the target artistic style feature data and the artistic style cluster feature data corresponding to the target artistic style type text feature data in the artistic style cluster feature data by using an intelligent optimization algorithm, searching for artistic style cluster feature data that matches the target artistic style feature data and performing data identification, thereby generating artistic style comparison data Q of the target artwork. mubiao ;

[0117] S711. Construct an artistic style feature to identify the water flow population, set the population size to N, the current number of iterations to t, and the maximum number of iterations to t. max The dimension of the search space for the art style clustering feature data is P;

[0118] Using the artistic style cluster feature data corresponding to the target artistic style type text feature data in the artistic style cluster feature data as an artistic style cluster feature data search space, randomly generating N artistic style cluster feature data in the artistic style cluster feature data search space, each artistic style cluster feature data corresponding to an artistic style feature identification water flow individual in the artistic style feature identification water flow population;

[0119] S712. Calculate the fitness value between each individual artistic style feature recognition water flow in the artistic style feature recognition water flow population and the target artistic style feature data, sort each individual artistic style feature recognition water flow in the artistic style feature recognition water flow population from largest to smallest according to the fitness value, and select the artistic style feature recognition water flow individual with the highest fitness value as the current optimal individual; the fitness value calculation formula is as follows:

[0120]

[0121] Among them, f irepresents the fitness value between the i-th artistic style feature recognition water flow individual in the artistic style feature recognition water flow population and the target artistic style feature data, m i represents the characteristic vector of the artistic style cluster feature data corresponding to the i-th artistic style feature identification water flow individual in the artistic style feature identification water flow population, n represents the characteristic vector of the target artistic style feature data, and φ represents a correction value;

[0122] S713, calculating the permeability coefficient α of each individual water flow identified by the artistic style feature in the water flow population identified by the artistic style feature in the search space of the artistic style cluster feature data i ; The calculation formula of permeability coefficient is as follows:

[0123]

[0124] Among them, α i represents the permeability coefficient of the i-th individual water flow identified by the artistic style feature in the water flow population, f max and f min Respectively represent the fitness values corresponding to the best position and the worst position in the water flow population identified by the artistic style feature;

[0125] Each individual water flow identified by the artistic style feature in the water flow population identified by the artistic style feature controls its own behavior and updates its position in the artistic style cluster feature data search space according to the permeability coefficient. The position update formula is as follows:

[0126]

[0127] in, represents the position of the i-th individual in the population of artistic style feature recognition water flow after the position is updated, M i and M j Respectively represent the current positions of the i-th and j-th art style feature identification water flow individuals in the art style feature identification water flow population, r1, r2 and r all represent random numbers uniformly distributed between (0,1), λ and η respectively represent the upper and lower limits of the art style clustering feature data search space, M best represents the position of the current optimal individual, β represents the water consumption coefficient that changes with the number of current iterations, ω t represents the adaptive step size factor, and Among them, ω max and ω min Respectively represent the maximum and minimum adaptive step sizes;

[0128] S714: randomly select an artistic style feature identification water flow individual from the artistic style feature identification water flow population according to probability, and update the position of the selected artistic style feature identification water flow individual in the artistic style cluster feature data search space according to the position of the current optimal individual; the position update formula is as follows:

[0129]

[0130] in, represents the position of the oth artistic style feature identification water flow individual randomly selected from the artistic style feature identification water flow population according to probability after the position update, and r3 and r4 both represent random numbers uniformly distributed between (0,1);

[0131] S715, calculating the fitness value of each individual artistic style feature identification water flow in the artistic style feature identification water flow population after the position is updated, if the fitness value of the individual artistic style feature identification water flow after the position is updated is greater than the original fitness value, replacing the original position with the new position; otherwise, retaining the original position;

[0132] Re-sorting the individual artistic style feature identification water flow individuals in the artistic style feature identification water flow population according to their fitness values from large to small, and selecting the artistic style feature identification water flow individual with the highest fitness value as the new current optimal individual;

[0133] S716: Determine whether the current number of iterations t is greater than or equal to the maximum number of iterations t max , if the current number of iterations t is greater than or equal to the maximum number of iterations t max , the current optimal individual is taken as the global optimal solution; otherwise, the current iteration number t is increased by 1, and the process returns to S713;

[0134] S72: Combine the target artistic style type text feature data, the target artistic style feature data set, and the artistic style comparison data to generate artistic style comparison identification data W={F mubiao ,C,Q mubiao}, pushing the art style comparison and identification data to the art style identification platform through a wireless communication network and displaying it on a display screen.

[0135] The second embodiment is as follows:

[0136] See also Figure 2, a multimodal art style comparison and identification system, comprising a genre category text feature data acquisition module, an art style feature acquisition scheme matching module, a target art style feature data acquisition module, a target art style feature data clustering module, an art style analysis model construction module, a target artwork style analysis module and an art style comparison data matching module;

[0137] The genre category text feature data acquisition module collects genre category text feature data of the target artwork online through the art style identification platform;

[0138] The artistic style feature acquisition scheme matching module performs character feature matching on the genre category text feature data and the feature acquisition schemes for works of different genres using a Sunday matching algorithm to generate an artistic style feature acquisition scheme for a target artwork;

[0139] The target art style feature data acquisition module selects different sensors according to the art style feature acquisition scheme to collect multimodal art style feature data of the target artwork to obtain target art style feature data;

[0140] The target art style feature data clustering module collects multimodal art style feature data of different artworks corresponding to the genre category text feature data online through the art style identification platform to generate an art style feature data set, and clusters the art style feature data set using an agglomerative hierarchical clustering algorithm to generate an art style cluster feature data matrix, and sets corresponding art style type text feature data;

[0141] The art style analysis model construction module trains and tests the initial SVM model based on the art style cluster feature data matrix and the art style type text feature data to obtain an art style analysis model;

[0142] The target artwork style analysis module inputs the target artwork style feature data into the art style analysis model to analyze the art style of the target artwork and generate target art style type text feature data;

[0143] The artistic style comparison data matching module matches the target artistic style feature data with the artistic style clustering feature data corresponding to the target artistic style type text feature data through an intelligent optimization algorithm, generates artistic style comparison data of the target artwork, and constructs artistic style comparison identification data, which is pushed to the artistic style identification platform through a wireless communication network and displayed on a display screen.

[0144] Those skilled in the art will appreciate that all or part of the steps in the above-mentioned embodiment methods can be accomplished by instructing the relevant hardware through a program. Therefore, the present application may take the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware. Furthermore, the present application may take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0145] Throughout this specification, references to terms such as "one embodiment," "example," or "specific example" indicate that the specific features, structures, materials, or characteristics described in conjunction with that embodiment or example are included in at least one embodiment or example of the invention. In this specification, schematic representations of these terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in any one or more embodiments or examples.

[0146] The preferred embodiments of the invention disclosed above are intended only to help illustrate the invention. These preferred embodiments do not exhaust all details, nor do they limit the invention to the specific embodiments described. Obviously, many modifications and variations are possible based on the content of this specification. These embodiments are selected and described in detail in this specification to better explain the principles and practical applications of the invention, thereby enabling those skilled in the art to better understand and utilize the invention.

Claims

1. A multimodal art style comparison and identification method, characterized in that: The steps include: S1. Collecting text feature data of the genre category of the target artwork; S2. Matching the genre category text feature data with feature acquisition schemes for works of different genres to generate an artistic style feature acquisition scheme for the target artwork; S3. Collecting artistic style feature data of a target artwork according to the artistic style feature acquisition scheme to obtain target artistic style feature data; S4, collecting artistic style feature data of different artworks corresponding to the genre category text feature data to obtain an artistic style feature data set, performing clustering processing on the artistic style feature data set to generate an artistic style cluster feature data matrix, and setting corresponding artistic style type text feature data; S5. Constructing an artistic style analysis model based on the artistic style cluster feature data matrix and the artistic style type text feature data; S6. Analyze the artistic style of the target artwork according to the artistic style analysis model and the target artistic style feature data to generate target artistic style type text feature data; S7: Match the target artistic style feature data with the artistic style cluster feature data corresponding to the target artistic style type text feature data to generate artistic style comparison data of the target artwork.

2. The multimodal artistic style comparison and identification method according to claim 1, characterized in that: The collection of genre category text feature data includes: The genre category text feature data A of the target artwork is collected online through the art style identification platform.

3. The multimodal artistic style comparison and identification method according to claim 2, characterized in that: Said S2 includes: S21. Establish a set of solutions for acquiring features of works of different genres in, represents the artistic style feature acquisition scheme corresponding to the i-th artwork genre, and k represents the total number of artwork genres; The genres of the artistic works include but are not limited to painting, sculpture, music, and film and television; The artistic style feature acquisition scheme represents the multimodal artistic style feature data required to be acquired for different genres of artworks; The multimodal art style feature data includes but is not limited to any one or more of image feature data, audio feature data, three-dimensional feature data, text feature data, and video feature data; S22, performing character feature matching on the genre category text feature data A and the different genre work feature acquisition schemes in the different genre work feature acquisition scheme set using the Sunday matching algorithm; If the genre text feature data A successfully matches any of the feature acquisition schemes for different genre works in the feature acquisition scheme set for different genre works, the character feature matching is terminated, and the feature acquisition schemes for different genre works that match the genre text feature data A are data identified to generate the artistic style feature acquisition scheme A for the target artwork. mubiao ; Otherwise, continue to perform character adjustment matching until the genre category text feature data A is successfully matched with any one of the feature acquisition schemes for different genre works in the feature acquisition scheme set for different genre works.

4. The multimodal artistic style comparison and identification method according to claim 3, characterized in that: Obtaining target artistic style feature data includes: Acquisition scheme A based on the artistic style features mubiao Different sensors are selected to collect multimodal artistic style feature data of the target artwork, and a target artistic style feature dataset B = {b1, b2, ..., b i ,…,b l }, where b i represents the i-th multimodal art style feature data of the target artwork, and l represents the total number of multimodal art style feature data of the target artwork.

5. The multimodal artistic style comparison and identification method according to claim 4, characterized in that: Said S4 includes: S41, collect multimodal art style feature data of different works of art corresponding to the genre category text feature data online through the art style identification platform, and generate an art style feature dataset C = {c1, c2, ..., c i ,…,c p }, where c i represents the artistic style feature data of the i-th artwork collected by the artistic style identification platform, and p represents the total number of artworks collected by the multimodal artistic style identification platform; S42, using an agglomerative hierarchical clustering algorithm to perform data clustering processing on the artistic style feature data in the artistic style feature data set to generate an artistic style cluster feature data matrix as follows: in, represents the multimodal art style feature data of the jth artwork contained in the i-th art style collected through the multimodal art style identification platform, d i represents the total number of artworks contained in the i-th category of artistic style collected through the multimodal artistic style identification platform; S43, annotating the corresponding art style category text feature data for each art style in the multimodal art style clustering feature data matrix, and generating an art style category text feature dataset E = {e1, e2, ..., e i ,…,e o }, where e i Represents the text feature data corresponding to the i-th artistic style.

6. The multimodal art style comparison and identification method according to claim 5, characterized in that: The data clustering process in S4 includes: S421: Treat each artistic style feature data in the artistic style feature data set as a cluster, calculate the distance between each cluster, and obtain a distance matrix. The distance calculation formula is as follows: Where D(x,y) represents the distance between vector x and vector y, x and y represent the l-dimensional feature vectors of any two cluster centers, x i Represents the value of vector x in the i-th dimension, y i Represents the value of vector y in the i-th dimension; S422, merging the two clusters corresponding to the minimum values in the distance matrix to obtain a new cluster, calculating the distance between the new cluster and other clusters, and updating the distance matrix according to the calculation result; S423, setting a distance threshold; If all distances in the distance matrix are greater than the distance threshold, the data clustering process is terminated to obtain the artistic style clustering feature data matrix. Otherwise, return to S422 until all distances in the distance matrix are greater than the distance threshold.

7. The multimodal art style comparison and identification method according to claim 6, characterized in that: Said S5 includes: S51, constructing an initial SVM model, setting a training data ratio, and performing data partitioning on the artistic style cluster feature data matrix and the artistic style type text feature data set according to the training data ratio to obtain an artistic style cluster feature training data matrix, an artistic style type text feature training data set, an artistic style cluster feature test data matrix, and an artistic style type text feature test data set; S52, setting a training error threshold and a maximum number of training times, inputting the artistic style cluster feature training data matrix and the artistic style type text feature training data set as training data and training label data into the initial SVM model for training, adjusting initial parameters of the initial SVM model according to the training results, and then continuing to train the SVM with the adjusted parameters until the training error is less than the training error threshold or the number of training times is greater than the maximum number of training times, thereby obtaining a trained SVM model; S53. Set a test accuracy threshold, input the artistic style cluster feature test data matrix and the artistic style type text feature test data set as test data and test label data respectively into the trained SVM model for testing, calculate the test accuracy, and if the test accuracy is greater than the test accuracy threshold, obtain the artistic style analysis model; otherwise, return to S52 and retrain until the test accuracy is greater than the test accuracy threshold.

8. The multimodal art style comparison and identification method according to claim 7, characterized in that: Generating target artistic style text feature data includes: The target artistic style feature data is input into the artistic style analysis model to analyze the artistic style of the target artwork, and generate target artistic style type text feature data F mubiao .

9. The multimodal artistic style comparison and identification method according to claim 8, characterized in that: The S7 includes: S71: Performing artistic style feature matching processing on the target artistic style feature data and the artistic style cluster feature data corresponding to the target artistic style type text feature data in the artistic style cluster feature data by using an intelligent optimization algorithm, searching for artistic style cluster feature data that matches the target artistic style feature data and performing data identification, thereby generating artistic style comparison data Q of the target artwork. mubiao ; S711. Construct an artistic style feature to identify the water flow population, set the population size to N, the current number of iterations to t, and the maximum number of iterations to t. max The dimension of the search space for the art style clustering feature data is P; Using the artistic style cluster feature data corresponding to the target artistic style type text feature data in the artistic style cluster feature data as an artistic style cluster feature data search space, randomly generating N artistic style cluster feature data in the artistic style cluster feature data search space, each artistic style cluster feature data corresponding to an artistic style feature identification water flow individual in the artistic style feature identification water flow population; S712. Calculate the fitness value between each individual artistic style feature recognition water flow in the artistic style feature recognition water flow population and the target artistic style feature data, sort each individual artistic style feature recognition water flow in the artistic style feature recognition water flow population from largest to smallest according to the fitness value, and select the artistic style feature recognition water flow individual with the highest fitness value as the current optimal individual; the fitness value calculation formula is as follows: Among them, f i represents the fitness value between the i-th artistic style feature recognition water flow individual in the artistic style feature recognition water flow population and the target artistic style feature data, m i represents the characteristic vector of the artistic style cluster feature data corresponding to the i-th artistic style feature identification water flow individual in the artistic style feature identification water flow population, n represents the characteristic vector of the target artistic style feature data, and φ represents a correction value; S713, calculating the permeability coefficient α of each individual water flow identified by the artistic style feature in the water flow population identified by the artistic style feature in the search space of the artistic style cluster feature data i ; The calculation formula of permeability coefficient is as follows: Among them, α i represents the permeability coefficient of the i-th individual water flow identified by the artistic style feature in the water flow population, f max and f min Respectively represent the fitness values corresponding to the best position and the worst position in the water flow population identified by the artistic style feature; Each individual water flow identified by the artistic style feature in the water flow population identified by the artistic style feature controls its own behavior and updates its position in the artistic style cluster feature data search space according to the permeability coefficient. The position update formula is as follows: in, represents the position of the i-th individual in the population of artistic style feature recognition water flow after the position is updated, M i and M j Respectively represent the current positions of the i-th and j-th art style feature identification water flow individuals in the art style feature identification water flow population, r1, r2 and r all represent random numbers uniformly distributed between (0,1), λ and η respectively represent the upper and lower limits of the art style clustering feature data search space, M best represents the position of the current optimal individual, β represents the water consumption coefficient that changes with the number of current iterations, ω t represents the adaptive step size factor, and Among them, ω max and ω min Respectively represent the maximum and minimum adaptive step sizes; S714: randomly select an artistic style feature identification water flow individual from the artistic style feature identification water flow population according to probability, and update the position of the selected artistic style feature identification water flow individual in the artistic style cluster feature data search space according to the position of the current optimal individual; the position update formula is as follows: in, represents the position of the oth artistic style feature identification water flow individual randomly selected from the artistic style feature identification water flow population according to probability after the position update, and r3 and r4 both represent random numbers uniformly distributed between (0,1); S715, calculating the fitness value of each individual artistic style feature identification water flow in the artistic style feature identification water flow population after the position is updated, if the fitness value of the individual artistic style feature identification water flow after the position is updated is greater than the original fitness value, replacing the original position with the new position; otherwise, retaining the original position; Re-sorting the individual artistic style feature identification water flow individuals in the artistic style feature identification water flow population according to their fitness values from large to small, and selecting the artistic style feature identification water flow individual with the highest fitness value as the new current optimal individual; S716: Determine whether the current number of iterations t is greater than or equal to the maximum number of iterations t max , if the current number of iterations t is greater than or equal to the maximum number of iterations t max , the current optimal individual is taken as the global optimal solution; otherwise, the current iteration number t is increased by 1, and the process returns to S713; S72: Combine the target artistic style type text feature data, the target artistic style feature data set, and the artistic style comparison data to generate artistic style comparison identification data W={F mubiao ,C,Q mubiao }, pushing the art style comparison and identification data to the art style identification platform through a wireless communication network and displaying it on a display screen.

10. A system for implementing the multimodal art style comparison and identification method according to any one of claims 1 to 9.

Images