Intelligent assessment feedback method and system for artistic creation

Abstract

According to the intelligent assessment feedback method and system for artistic creation, the artistic work data uploaded by the user is obtained, the visual perception features and semantic features of the artistic work data are fused and extracted, and the limitation of feature association missing caused by traditional single-mode analysis is broken through; different modal features are dynamically associated by using a cross-modal attention mechanism, and an evaluation result covering multiple dimensions such as color, composition, style and the like is generated, so that the problems of fixed evaluation dimension and poor work type adaptability in the prior art are solved; and comprehensive feedback information containing structured evaluation and personalized recommendation is generated based on a multi-dimensional evaluation result, and a closed loop from data analysis to creation guidance is realized.

Description

Technical Field

[0001] The present invention belongs to the field of computer science and artificial intelligence technology, and in particular relates to an intelligent evaluation and feedback method for artistic creation. Background Art

[0002] With the development of artificial intelligence and computer vision, deep learning-based art evaluation technology has emerged. This technology uses algorithmic models to automatically analyze the visual features of artworks, such as color distribution and line structure, partially replacing traditional manual evaluation. This technology is characterized by its reliance on data-driven feature extraction and attempts to provide objective evaluation criteria through quantitative metrics such as hue deviation and composition entropy. Traditionally, the evaluation of artistic creations has relied primarily on human expert experience, such as professional artists or critics who subjectively score works based on aesthetic theory, such as color matching, compositional balance, and stylistic consistency. In the early stages of automated technology, some tools attempted to adopt single-modal analysis methods, such as analyzing color distribution solely through image histograms or using edge detection algorithms to extract simple compositional features. However, current automated evaluation methods have significant problems: Insufficient multimodal feature fusion: Traditional methods often process visual features and semantic features in isolation, lacking cross-modal associations such as the linkage analysis of color coordination and style classification, resulting in a single evaluation dimension; Dynamic adaptability defects: Existing algorithms usually use fixed weights such as static scoring ratios for color, composition, and style, and cannot dynamically adjust evaluation strategies according to the type of work such as oil painting, watercolor, digital art, etc.; Low practicality of feedback information: Traditional automated systems only provide numerical scores or simple labels such as composition imbalance, and lack structured improvement suggestions and personalized creation guidance based on evaluation results; Insufficient style matching accuracy: Early style classification models relied on shallow features such as brushstroke direction statistics, and it was difficult to capture high-order style tensors, resulting in large similarity calculation errors. Summary of the Invention

[0003] Based on this, it is necessary to provide an intelligent evaluation and feedback method for artistic creation that can solve the above problems.

[0004] In a first aspect, the present application provides an intelligent evaluation and feedback method for artistic creation, comprising:

[0005] Get artwork data uploaded by users;

[0006] Extract multimodal feature information including visual perception features and semantic features from artwork data;

[0007] Based on multimodal feature information, a cross-modal attention mechanism is used to generate multi-dimensional evaluation results;

[0008] Generate feedback information based on multi-dimensional evaluation results.

[0009] In one embodiment, multimodal feature information including visual perception features and semantic features is extracted from art work data, including:

[0010] Preprocess the art work data to obtain standardized data;

[0011] Use a convolutional neural network to process the standardized data to obtain color distribution features;

[0012] Analyze the standardized data through an edge detection algorithm to obtain line structure features;

[0013] Classify the standardized data based on a transfer learning model to obtain style classification features.

[0014] In one embodiment, according to the multimodal feature information, a multi-dimensional evaluation result is generated using a cross-modal attention mechanism, including:

[0015] According to the color distribution features in the feature information, use a hue ring matching function to generate a color evaluation result;

[0016] According to the line structure features in the feature information, use a visual focus heat map to generate a composition evaluation result;

[0017] According to the style classification features in the feature information, perform similarity matching using a preset work feature database to generate a style evaluation result.

[0018] In one embodiment, feedback information is generated according to the multi-dimensional evaluation result, including:

[0019] According to the art type to which the art work data belongs, use a multi-layer perceptron algorithm to determine the weight coefficients of each multi-dimensional evaluation result;

[0020] Use the weight coefficients to perform weighted processing on the color evaluation result, composition evaluation result, and style evaluation result to obtain comprehensive evaluation information;

[0021] Generate an evaluation report including a multi-level scoring system according to the comprehensive evaluation information.

[0022] In one embodiment, use the weight coefficients to perform weighted processing on the color evaluation result, composition evaluation result, and style evaluation result to obtain comprehensive evaluation information, including:

[0023] Calculate the comprehensive evaluation index using the following formula:

[0024]

[0025] Among them, S represents the comprehensive evaluation index, represents the color coordination degree, ΔH irepresents the hue deviation value of the i-th color patch, μ H represents the ideal hue mean value, σ H represents the adaptive threshold represents the composition entropy value, P(x,y) represents the attention probability of the coordinate point represents the proportion of the area of the region of interest represents the style similarity, F represents the tensor norm, τ c represents the third-order feature tensor of the work represents the k-th reference style tensor represents the dynamic weight coefficient, λ represents the eigenvalue of the covariance matrix of the work type, θ represents the non-linear adjustment factor represents the style difference degree represents the difference degree reward function

[0026] In one embodiment, feedback information is generated according to the multi-dimensional evaluation results, including:

[0027] Using a natural language processing model to process the multi-dimensional evaluation results to generate a structured evaluation text

[0028] Determine the type of creation goal according to the style evaluation result in the structured evaluation text

[0029] According to the type of creation goal, use the collaborative filtering algorithm to obtain recommended works from the preset reference work library and generate personalized recommendation information

[0030] In one embodiment, using a natural language processing model to process the multi-dimensional evaluation results to generate a structured evaluation text, including:

[0031] Using the semantic parsing layer of the natural language processing model to extract the key evaluation features in the multi-dimensional evaluation results, and the key evaluation features are determined based on the type of creation goal

[0032] Based on the key evaluation features and the preset evaluation template, use the text generation layer of the natural language processing model to generate a structured evaluation text including the technical dimension evaluation description, style dimension evaluation and corresponding improvement suggestions

[0033] In a second aspect, the present application also provides an intelligent evaluation feedback system for artistic creation, including:

[0034] A data acquisition module for obtaining the artistic work data uploaded by the user

[0035] A multi-modal feature parsing module for extracting multi-modal feature information including visual perception features and semantic features from the artistic work data

[0036] An intelligent evaluation and decision-making module, which is used to generate multi-dimensional evaluation results by using a cross-modal attention mechanism according to multi-modal feature information;

[0037] A feedback generation and recommendation module, which is used to generate feedback information according to multi-dimensional evaluation results.

[0038] In a third aspect, the present application also provides a computer device, including a memory and a processor. The memory stores a computer program, and when the processor executes the computer program, the steps of the above-mentioned intelligent evaluation and feedback method for artistic creation are implemented:

[0039] In a fourth aspect, the present application also provides a computer-readable storage medium, on which a computer program is stored. When the computer program is executed by a processor, the steps of the above-mentioned intelligent evaluation and feedback method for artistic creation are implemented:

[0040] The above-mentioned intelligent evaluation and feedback method and system for artistic creation construct a multi-modal information matrix by simultaneously extracting visual perception features and semantic features in an artistic work, and use a cross-modal attention mechanism to break through the bottleneck of the lack of feature correlation caused by traditional single-modal analysis; based on multi-modal feature information, multi-dimensional evaluation results covering color, composition, style, etc. are dynamically generated through cross-modal attention weights, overcoming the problem of poor adaptability of traditional fixed evaluation dimensions and static evaluation criteria to different types of works; the multi-dimensional evaluation results are transformed into structured feedback information to realize a closed loop from data analysis to creation guidance. BRIEF DESCRIPTION OF THE DRAWINGS

[0041] In order to more clearly illustrate the technical solutions in the embodiments of the present application or related technologies, the following will briefly introduce the drawings required for use in the description of the embodiments or related technologies. Obviously, the drawings in the following description are only some embodiments of the present application. For those of ordinary skill in the art, other drawings can be obtained based on these drawings without creative efforts.

[0042] Figure 1 It is a flowchart of an intelligent evaluation and feedback method for artistic creation according to the present invention;

[0043] Figure 2 It is a structural diagram of an intelligent evaluation and feedback system for artistic creation according to the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0044] In order to make the purpose, technical solutions and advantages of the present application clearer, the following will further describe the present application in detail with reference to the drawings and embodiments. It should be understood that the specific embodiments described here are only used to explain the present application, and are not used to limit the present application.

[0045] The intelligent evaluation and feedback method and system for artistic creation of the present application, the related hardware architecture includes a data acquisition device, a user terminal, and a computing server. When a user has a need for artistic creation evaluation, the data acquisition device is used to obtain artistic work data and upload it to the server through the network. The server performs feature extraction, evaluation, and feedback generation, and finally sends the results to the user terminal.

[0046] In one embodiment, as Figure 1 shown, an intelligent evaluation and feedback method for artistic creation is provided. In this embodiment, this method is described by taking its application to a collaborative system of a data acquisition device, a user terminal, and a computing server as an example. It can be understood that this method can also be applied alone to a user terminal integrating data acquisition and computing functions, and can also be applied to a system including a user terminal with integrated data acquisition and a computing server, and is realized through the interaction between the user terminal and the computing server. In this embodiment, the method includes the following steps:

[0047] S101, Obtain the artistic work data uploaded by the user.

[0048] Among them, the artistic work data can be obtained by means such as photographing the artistic work with a camera or scanning it with a scanning instrument, and the artistic work data is uploaded to the server through the network. This process realizes the transfer of the artistic work from offline to online, from the user side to the system processing side. The types of artistic work data can include, but are not limited to, paintings, photographs, digital art works, etc. The artistic information that the artistic work data can contain are: visual perception features such as picture composition, color distribution, line structure, and light and shadow effects, as well as semantic features contained in the artistic work, such as the cultural connotations conveyed by the work style, etc., which are the basis for subsequent multi-modal feature extraction.

[0049] S102, Extract multi-modal feature information including visual perception features and semantic features from the artistic work data.

[0050] Among them, visual perception features and semantic features can reflect the characteristics of the work from different angles. By extracting visual perception features and semantic features, a comprehensive work information matrix can be constructed to avoid the limitations of single-modal analysis. It can be to preprocess the artistic work data to obtain standardized data, make the data format unified and the specifications consistent, and then use a convolutional neural network to process the standardized data to obtain color distribution features; use an edge detection algorithm to analyze the standardized data to obtain line structure features; classify the standardized data based on a transfer learning model to obtain style classification features. Through multi-modal feature information extraction, the bottleneck of the lack of feature correlation caused by traditional single-modal analysis is broken.

[0051] S103, According to the multi-modal feature information, use a cross-modal attention mechanism to generate multi-dimensional evaluation results.

[0052] The cross-modal attention mechanism dynamically correlates visual perception features with semantic features, exploring potential connections between features of different modalities. For example, color harmony and style classification features are not isolated but rather interrelated, influencing the evaluation results. Extracting these connections through the cross-modal attention mechanism can make the evaluation more consistent with the actual situation of the artwork. This can be done by judging the rationality and harmony of color matching based on color distribution characteristics; analyzing the composition of the picture based on line structure characteristics; and determining the similarity and uniqueness of the work's style to classic styles based on style classification characteristics. By combining these features to explore their mutual influence and generate multi-dimensional evaluation results, users can understand the strengths and weaknesses of the work from multiple dimensions such as color, composition, and style, achieving a closed loop from data analysis to creative guidance.

[0053] S104: Generate feedback information based on the multi-dimensional evaluation results.

[0054] Different art genres place varying emphasis on aspects such as color, composition, and style. For example, oil paintings prioritize color expression, while sketches emphasize line structure. Based on the multi-dimensional evaluation results of an artwork, its genre can be determined, and the weighting of each dimension can be further determined. Based on these weightings, mathematical models, such as natural language processing models, can be used to process the multi-dimensional evaluation results and generate feedback tailored to the genre. This feedback can be structured text, allowing users to clearly understand the work's scores across different dimensions and its overall performance.

[0055] The intelligent evaluation and feedback method and system of artistic creation in the present application extracts multimodal feature information containing visual perception features and semantic features in the artwork, which breaks the limitations of traditional isolated processing of visual and semantic features, so that the evaluation is no longer limited to a single modality, and can reflect the characteristics of the work from multiple angles. The cross-modal attention mechanism is used to generate multi-dimensional evaluation results, overcoming the problem of solidification of traditional evaluation dimensions. Dynamic evaluation can be performed according to the characteristics of different works, improving adaptability to different types of works. Feedback information is generated based on the multi-dimensional evaluation results to avoid the problem of low practicality of feedback information in traditional automated systems. It is not just a simple numerical score or label, but provides content with more practical guidance. Therefore, this method can solve the problems of insufficient multimodal feature fusion, dynamic adaptability defects, low practicality of feedback information, and insufficient style matching accuracy in the technical background, thereby improving the quality and effect of artistic creation evaluation and providing users with more valuable references.

[0056] In one embodiment, extracting multimodal feature information including visual perception features and semantic features from artwork data includes:

[0057] S201, preprocessing artwork data to obtain standardized data;

[0058] S202. Process the standardized data using a convolutional neural network to obtain color distribution features;

[0059] S203. Analyze the standardized data through an edge detection algorithm to obtain line structure features;

[0060] S204. Classify the standardized data based on a transfer learning model to obtain style classification features.

[0061] Specifically, the data sources of artworks are diverse, with differences in formats, resolutions, color modes, etc. Through preprocessing, the data format can be unified, the image size can be adjusted, the color values can be normalized, etc., so that the data meets the requirements of subsequent processing models, improving the accuracy and efficiency of analysis. A convolutional neural network is good at automatically extracting local features in an image. Through structures such as convolutional layers, pooling layers, and fully connected layers, it can effectively capture information such as the distribution law of colors, contrast, and color hierarchy in the image, providing a basis for evaluating the color application of the work, such as judging whether the color combination is harmonious and whether it highlights the theme. The edge detection algorithm can identify the edges of objects in the image and determine information such as the position, direction, and length of the lines. By analyzing the line structure, the composition layout of the work, the shape and spatial relationship of the objects, etc., such as the stability of the picture and the visual guiding effect, can be evaluated, helping to judge the advantages and disadvantages of the work in terms of composition. The transfer learning model can use the knowledge pre-trained on a large-scale image dataset to quickly and accurately classify the style of artworks, identify which style the work belongs to, such as realistic, abstract, impressionist, etc., and can also discover the unique features and improvement directions of the work style by comparing with known style works.

[0062] In one embodiment, according to the multi-modal feature information, use a cross-modal attention mechanism to generate multi-dimensional evaluation results, including:

[0063] S301. According to the color distribution features in the feature information, use a hue ring matching function to generate a color evaluation result;

[0064] S302. According to the line structure features in the feature information, use a visual focus heat map to generate a composition evaluation result;

[0065] S303. According to the style classification features in the feature information, perform similarity matching using a preset work feature database to generate a style evaluation result.

[0066] Exemplarily, for the color distribution features extracted in the previous step, a hue ring matching function can be used to analyze the coordination of color combinations. The hue ring matching function is based on color theories such as complementary colors and analogous color matching principles, quantifies the difference between the hue deviation of different color patches and the ideal mean value, and generates a color evaluation result, such as judging whether the colors in the picture are harmonious, whether there are visual conflicts or monotony problems; for the line structure features, the distribution path of the audience's line of sight in the picture can be visualized by means of a visual focus heat map. By calculating the attention probability of each coordinate point and the proportion of the area of the region of interest, the balance of the composition, the visual guidance effect, and the degree of focus prominence are evaluated, such as judging whether the main body is clear, whether the composition has a sense of dynamics or stability; for the style classification features, the third-order feature tensors of the art work and the reference style works in the preset work feature database can be matched for similarity. By calculating the cosine similarity through the tensor norm, a style evaluation result is generated, which can judge the similarity degree of the style of the art work to a certain art genre such as Impressionism or Abstractism, or its performance in style innovation, so as to achieve accurate quantitative evaluation of color, composition, and style.

[0067] In one embodiment, feedback information is generated according to the multi-dimensional evaluation results, including:

[0068] S401, according to the art type to which the art work data belongs, use the multi-layer perceptron algorithm to determine the weight coefficients of each multi-dimensional evaluation result;

[0069] S402, use the weight coefficients to perform weighted processing on the color evaluation result, the composition evaluation result, and the style evaluation result to obtain comprehensive evaluation information;

[0070] S403, according to the comprehensive evaluation information, generate an evaluation report including a multi-level scoring system.

[0071] Specifically, according to the type of art works, such as oil paintings, watercolors, digital arts, etc., the multi-layer perceptron algorithm can be used to analyze the feature distributions of different types of works in historical data, and dynamically determine the weight coefficients of multi-dimensional evaluation results such as color, composition, and style. For example, oil painting creation may pay more attention to color levels, and the color evaluation weight is higher; photographic works may rely more on the visual impact of composition, and the composition evaluation weight is higher. By learning the typical features of different types of works, the model can adaptively adjust the weight coefficients, avoiding the use of fixed static scoring ratios in traditional methods, and improving the pertinence of evaluation to the work type. After weighted calculation and processing, the overall performance of the work in aspects such as color harmony, composition rationality, and style matching degree can be quantified. The weighting process takes into account the core demands of different art types, avoids a single dimension dominating the evaluation, and makes the results more comprehensive. Based on the comprehensive evaluation information, a report containing a multi-level scoring system can be generated, which can be divided into the following four-level scoring contents. For example, the first-level indicators: the individual scores of color, composition, and style, such as 85 points for color, 78 points for composition, and 90 points for style; the second-level indicators: the detailed evaluations of each dimension, such as the hue balance degree and contrast rationality under the color dimension; the comprehensive score: the overall score obtained through weighted calculation, such as the comprehensive score of 86 points; the index visualization charts: using color distribution histograms, composition heat maps, style similarity radar charts, etc., to assist users in intuitively understanding the advantages and disadvantages of the works. Compared with traditional systems that only provide a single numerical score or a simple label such as poor composition, the multi-level scoring system presents the evaluation results in a hierarchical structure, retaining both details and reflecting the whole, providing users with clear improvement directions, such as the composition entropy value is relatively low, and the visual focus guidance needs to be enhanced. It can solve the problems of dynamic adaptability defects and low feedback practicality in traditional technologies.

[0072] In one embodiment, the weight coefficients are used to perform weighted processing on the color evaluation result, the composition evaluation result, and the style evaluation result to obtain comprehensive evaluation information, including:

[0073] S501, calculate the comprehensive evaluation index using the following formula:

[0074]

[0075] where S represents the comprehensive evaluation index, represents the color harmony degree, ΔH i represents the hue deviation value of the i-th color block, μ H represents the ideal hue mean value, σ H represents the adaptive threshold, represents the composition entropy value, P(x,y) represents the attention probability of the coordinate point, represents the area ratio of the interest area, represents the style similarity, F represents the tensor norm, τ cRepresents the third-order feature tensor of the work, Represents the k-th reference style tensor, Represents the dynamic weight coefficient, λ represents the eigenvalue of the covariance matrix of the work type, and θ represents the non-linear adjustment factor, Represents the style difference degree, Represents the difference degree reward function.

[0076] Specifically, the color coordination degree Based on the variance standardization in statistics, calculate the actual hue deviation ΔH i And the ideal mean value μ H The normalized difference. If the variance is smaller, that is, the color coordination degree C is closer to 1, it means that the color coordination is better. The composition entropy value Draws on the information entropy theory, multiplies the coordinate point attention probability P(x, y) by the proportion of the area of the region of interest To evaluate the visual guidance effect of the composition. A low entropy value represents a prominent main body, and a high entropy value may indicate a scattered composition. The style similarity Calculates the cosine similarity τ using the inner product and norm of tensors c Represents the third-order feature tensor of the work, Represents the k-th reference style tensor. When the calculated similarity in the matching is the maximum value maxS sim It means that it is closest to this style, and the size of the maximum value represents the degree of closeness to this style. The dynamic weight coefficient Is used to reflect the weight preference of different art types and solve the defect of traditional fixed weights. The non-linear adjustment factor θ is used to adjust the correlation strength between dimensions through power operations. For example, when θ > 1, it amplifies the influence of the dominant dimension, such as highlighting style innovation. When θ < 1, it balances the contributions of each dimension to avoid a single dimension dominating the evaluation. The style difference degree Is used to calculate the average difference between the art work and all reference styles. The larger the value of D, the more unique the style or the more deviated from the preset style. The difference degree reward function is When D > τ, it means that the style difference is large, such as an innovative style, and the reward term increases superlinearly with the increase of D, encouraging style innovation; when D ≤ τ, it means that the style is close to the preset, such as a realistic style, and slightly punishes the lack of difference to ensure style consistency. Through the data-driven weight allocation and non-linear reward mechanism, it breaks through the limitation of the traditional fixed weight + linear superposition in evaluation and realizes: evaluation adaptability, dynamically adjusts the importance of evaluation dimensions for different types of works such as oil paintings and photography; innovation incentive, balances style inheritance and innovation through the style difference reward term; scientific quantification, transforms subjective art evaluation into a computable mathematical model, combines statistics, information theory and tensor analysis, and improves the objectivity and interpretability of evaluation.

[0077] In one embodiment, feedback information is generated according to the multi-dimensional evaluation result, including:

[0078] S601, using a natural language processing model to process the multi-dimensional evaluation result to generate a structured evaluation text;

[0079] S602, determining the type of creation goal according to the style evaluation result in the structured evaluation text;

[0080] S603, according to the type of creation goal, using a collaborative filtering algorithm to obtain recommended works from a preset reference work library to generate personalized recommendation information.

[0081] Exemplarily, the natural language processing model has semantic understanding and text generation capabilities, and can analyze and interpret multi-dimensional data such as the color evaluation result, composition evaluation result, and style evaluation result obtained in the foregoing steps, and convert them into natural language descriptions that are easy for humans to understand. The model will organize language and generate structured evaluation text according to preset rules and templates, combined with the specific values and features in the multi-dimensional evaluation results. For example, for the color evaluation result, if the color coordination degree is relatively high, it may generate a description such as "the color matching of the work is very harmonious, the different colors complement each other, creating a comfortable visual experience"; if the composition entropy value is relatively low, indicating that the composition focus is concentrated, it may be expressed as "the picture composition is simple and clear, the main body is prominent, and it can quickly attract the audience's attention". It provides users with intuitive and detailed work evaluation feedback, enabling users to clearly understand the advantages and disadvantages of the work in each dimension without the need for professional data analysis knowledge. In the structured evaluation text, information such as the style characteristics of the work, the similarity to a certain style, and the unique features of the style can be described through the evaluation part about the style, and the current style result of the work can be judged. According to the style evaluation result, combined with the user's intention or the development direction of the work itself, the creation target type can be determined. For example, if the style of the work is close to a certain traditional style, but the user hopes to innovate, then the creation target type may be the traditional style innovation type; if the work style is relatively vague and lacks a clear positioning, the creation target type may be the clear style guidance type. Clearly defining the creation target type provides a direction for subsequent recommendation of appropriate reference works, making the recommendation more targeted. The collaborative filtering algorithm is a recommendation algorithm based on user behavior and data. By analyzing the creation target type and the characteristics of the works in the preset reference work library, works similar to the creation target type are found. The similarity between different works is calculated, and the works with higher similarity are more likely to be recommended. According to the calculation results, works with a higher matching degree to the creation target type are selected from the reference work library. These works may be similar to the user's goal in terms of style, theme, expression techniques, etc. The recommended works are organized into personalized recommendation information, which can include the name, author, brief description, and recommendation reasons of the works. The recommendation information can explain in which aspects the work fits the user's creation goal. For example, "this work has made bold color innovations on the basis of the traditional style, which is in line with your traditional style innovation type creation goal and is worth learning from", providing users with valuable work cases for reference. Users can obtain inspiration from these recommended works, learn excellent creation techniques and skills, and thus improve the quality of their own works.

[0082] In one embodiment, a natural language processing model is used to process the multi-dimensional evaluation results to generate a structured evaluation text, including:

[0083] S701. Extract key evaluation features from the multi-dimensional evaluation results using the semantic parsing layer of the natural language processing model. The key evaluation features are determined based on the type of creation goal.

[0084] S702. Based on the key evaluation features and a preset evaluation template, use the text generation layer of the natural language processing model to generate a structured evaluation text including a technical dimension evaluation description, a style dimension evaluation, and corresponding improvement suggestions.

[0085] Specifically, use the semantic parsing layer of the natural language processing model to screen out the core features strongly related to the type of creation goal from the multi-dimensional evaluation results such as color, composition, and style. Among them, feature selection is dynamically screened based on the user's creation goals such as realistic style, abstract expression, commercial illustration, etc. For example: If the creation goal is a realistic oil painting, the key features may focus on color coordination, line structure clarity, and light and shadow restoration; if the goal is abstract art, it focuses on style difference, composition innovation, and color emotional expression. Avoid the problem of information overload in traditional feedback and only retain the core evaluation points related to the user's goal; use the creation goal to anchor the features to make the feedback more in line with the actual needs of the user, such as students focusing on technical details and professional painters paying attention to style innovation. The preset evaluation template can include three core modules, such as a technical dimension evaluation description, a style dimension evaluation, and corresponding improvement suggestions. It can be to associate the key features with the template framework through the attention mechanism, convert numerical indicators into natural language descriptions, and generate professional and easy-to-understand suggestions by combining domain knowledge such as art theory and creation techniques. Avoid the defects of scattered labels in traditional feedback, present the evaluation conclusion in a clear hierarchy such as technology → style → suggestion, and the improvement suggestions are derived from quantitative indicators, such as the specific shortcoming of the C value corresponding to the single color level, rather than vague evaluation, and users can directly refer to and adjust;

[0086] Convert the output of the evaluation algorithm into human-readable natural language, taking into account both professionalism and usability, and applicable to different scenarios such as art education and professional creation.

[0087] The intelligent evaluation and feedback method and system for artistic creation of the present application achieve a comprehensive upgrade of artistic creation evaluation by using multi-dimensional technological innovation: extracting visual perception and semantic features through convolutional neural networks, edge detection algorithms, and transfer learning models, constructing a multi-modal information matrix, solving the problem of missing feature correlation in traditional single-modal analysis, and realizing multi-dimensional analysis of the color, composition, and style of works; using a cross-modal attention mechanism to dynamically associate different modal features, combining hue ring matching, visual focus heat map, and style tensor similarity matching to generate multi-dimensional evaluation results covering color coordination, composition rationality, and style uniqueness, breaking through the bottlenecks of fixed evaluation dimensions and poor adaptability to work types; dynamically determining weight coefficients based on a multi-layer perceptron, constructing a comprehensive evaluation model through non-linear adjustment and difference reward functions, realizing differential evaluation of different types of works such as oil paintings and photography, and balancing style inheritance and innovation incentives; generating structured text containing technical dimension descriptions, style evaluations, and improvement suggestions by means of natural language processing, and providing personalized recommended works through collaborative filtering, solving the problem of low practicality of traditional feedback, and forming a complete closed loop from data analysis to creation guidance; the system modular design realizes the efficient collaboration of data collection, feature analysis, evaluation decision-making, and feedback generation, providing an objective and interpretable intelligent evaluation solution for scenarios such as art education and professional creation, and significantly improving the scientificity and practicality of artistic creation evaluation.

[0088] It should be understood that although the steps in the flowcharts involved in the above-described embodiments are shown in sequence according to the arrows, these steps are not necessarily executed in the order indicated by the arrows. Unless there is a clear indication in this article, the execution of these steps has no strict order limit, and these steps can be executed in other orders. Moreover, at least a part of the steps in the flowcharts involved in the above-described embodiments may include multiple steps or multiple stages. These steps or stages are not necessarily executed at the same moment, but can be executed at different moments, and the execution order of these steps or stages is not necessarily sequential, but can be executed alternately or alternately with at least a part of other steps or steps or stages in other steps.

[0089] Based on the same inventive concept, the embodiments of the present application also provide an intelligent evaluation and feedback system for artistic creation for implementing the intelligent evaluation and feedback method for artistic creation involved above. The implementation solutions for solving problems provided by this system are similar to the implementation solutions recorded in the above method. Therefore, the specific limitations in one or more embodiments of the intelligent evaluation and feedback system for artistic creation provided below can refer to the limitations on the intelligent evaluation and feedback method for artistic creation in the above text, and will not be repeated here.

[0090] In an exemplary embodiment, as Figure 2As shown in the figure, an intelligent evaluation and feedback system for artistic creation is provided, including:

[0091] A data acquisition module 11, which is used to obtain the data of the artistic works uploaded by users;

[0092] A multi-modal feature parsing module 12, which is used to extract multi-modal feature information including visual perception features and semantic features from the artistic work data;

[0093] An intelligent evaluation and decision-making module 13, which is used to generate multi-dimensional evaluation results by using a cross-modal attention mechanism according to the multi-modal feature information;

[0094] A feedback generation and recommendation module 14, which is used to generate feedback information according to the multi-dimensional evaluation results.

[0095] In one embodiment, the multi-modal feature parsing module 12 is further used for:

[0096] Preprocess the artistic work data to obtain standardized data;

[0097] Process the standardized data by using a convolutional neural network to obtain color distribution features;

[0098] Analyze the standardized data through an edge detection algorithm to obtain line structure features;

[0099] Classify the standardized data based on a transfer learning model to obtain style classification features.

[0100] In one embodiment, the multi-modal feature parsing module 12 is further used for:

[0101] According to the color distribution features in the feature information, use a hue ring matching function to generate a color evaluation result;

[0102] According to the line structure features in the feature information, use a visual focus heat map to generate a composition evaluation result;

[0103] According to the style classification features in the feature information, perform similarity matching by using a preset work feature database to generate a style evaluation result.

[0104] In one embodiment, the feedback generation and recommendation module 14 is further used for:

[0105] According to the art type to which the artistic work data belongs, use a multi-layer perceptron algorithm to determine the weight coefficients of each multi-dimensional evaluation result;

[0106] Use the weight coefficients to perform weighted processing on the color evaluation result, the composition evaluation result and the style evaluation result to obtain comprehensive evaluation information;

[0107] Generate an evaluation report containing a multi-level scoring system based on the comprehensive evaluation information.

[0108] In one embodiment, the feedback generation and recommendation module 14 is further configured to:

[0109] Calculate the comprehensive evaluation index using the following formula:

[0110]

[0111] where S represents the comprehensive evaluation index, represents the color coordination degree, and ΔH i represents the hue deviation value of the i-th color patch, and μ H represents the ideal hue mean value, and σ H represents the adaptive threshold, represents the composition entropy value, P(x, y) represents the attention probability of the coordinate point, represents the proportion of the area of the region of interest, represents the style similarity degree, F represents the tensor norm, and τ c represents the third-order feature tensor of the work, represents the k-th reference style tensor, represents the dynamic weight coefficient, λ represents the eigenvalue of the covariance matrix of the work type, and θ represents the nonlinear adjustment factor, represents the style difference degree, represents the difference degree reward function.

[0112] In one embodiment, the feedback generation and recommendation module 14 is further configured to:

[0113] Process the multi-dimensional evaluation results using a natural language processing model to generate a structured evaluation text;

[0114] Determine the type of creation goal based on the style evaluation result in the structured evaluation text;

[0115] According to the type of creation goal, use a collaborative filtering algorithm to obtain recommended works from a preset reference work library and generate personalized recommendation information.

[0116] In one embodiment, the feedback generation and recommendation module 14 is further configured to:

[0117] Use the semantic analysis layer of the natural language processing model to extract the key evaluation features in the multi-dimensional evaluation results, and the key evaluation features are determined based on the type of creation goal;

[0118] Based on key evaluation features and a preset evaluation template, use the text generation layer of the natural language processing model to generate a structured evaluation text including technical dimension evaluation descriptions, style dimension evaluations, and corresponding improvement suggestions.

[0119] In one embodiment, a computer device is provided, including a memory and a processor. The memory stores a computer program, and when the processor executes the computer program, the steps of an intelligent evaluation and feedback method for artistic creation as described above are implemented.

[0120] In one embodiment, a computer-readable storage medium is provided, on which a computer program is stored. When the computer program is executed by a processor, the steps in the above method embodiments are implemented.

[0121] For the device embodiments, since they basically correspond to the method embodiments, the relevant parts can be referred to the partial descriptions of the method embodiments. The device embodiments described above are only illustrative. The components described as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed to multiple network units. Some or all of the modules can be selected according to actual needs to achieve the purpose of the present disclosure solution. Those of ordinary skill in the art can understand and implement it without creative efforts.

[0122] The above embodiments only represent several implementation manners of the embodiments of the present application. The descriptions are relatively specific and detailed, but should not be construed as a limitation on the patent scope of the application embodiments. It should be noted that for those of ordinary skill in the art, without departing from the concept of the embodiments of the present application, several modifications and improvements can still be made, and these all belong to the protection scope of the embodiments of the present application.

Claims

1. An intelligent evaluation and feedback method for artistic creation, characterized in that The method includes: Obtaining the artistic work data uploaded by the user; Extracting multimodal feature information including visual perception features and semantic features from the artistic work data; Generating multi-dimensional evaluation results by using a cross-modal attention mechanism according to the multimodal feature information; Generating feedback information according to the multi-dimensional evaluation results.

2. The method according to claim 1, characterized in that The extracting multimodal feature information including visual perception features and semantic features from the artistic work data includes: Preprocessing the artistic work data to obtain standardized data; Processing the standardized data by using a convolutional neural network to obtain color distribution features; Analyzing the standardized data through an edge detection algorithm to obtain line structure features; Classifying the standardized data based on a transfer learning model to obtain style classification features.

3. The method according to claim 1, wherein The generating multi-dimensional evaluation results by using a cross-modal attention mechanism according to the multimodal feature information includes: Generating a color evaluation result by using a hue ring matching function according to the color distribution features in the feature information; Generating a composition evaluation result by using a visual focus heat map according to the line structure features in the feature information; Performing similarity matching by using a preset work feature database according to the style classification features in the feature information to generate a style evaluation result.

4. The method according to claim 3, characterized in that, The generating feedback information according to the multi-dimensional evaluation results includes: Determining the weight coefficients of the multi-dimensional evaluation results by using a multi-layer perceptron algorithm according to the art type to which the artistic work data belongs; Performing weighted processing on the color evaluation result, the composition evaluation result, and the style evaluation result by using the weight coefficients to obtain comprehensive evaluation information; Generating an evaluation report including a multi-level scoring system according to the comprehensive evaluation information.

5. The method according to claim 4, wherein The performing weighted processing on the color evaluation result, the composition evaluation result, and the style evaluation result by using the weight coefficients to obtain comprehensive evaluation information includes: Calculating a comprehensive evaluation index by using the following formula: Among them, S represents the comprehensive evaluation index, represents the color coordination degree, ΔH i represents the hue deviation of the i-th color block, μ H represents the ideal hue mean value, σ H represents the adaptive threshold, represents the composition entropy value, P(x, y) represents the attention probability of the coordinate point, represents the proportion of the area of the region of interest, represents the style similarity, F represents the tensor norm, τ c represents the third-order feature tensor of the work represents the k-th reference style tensor, represents the dynamic weight coefficient, λ represents the eigenvalue of the covariance matrix of the work type, θ represents the nonlinear adjustment factor, represents the style difference degree, represents the difference degree reward function.

6. The method according to claim 1, characterized in that, The generating feedback information according to the multi-dimensional evaluation results includes: Processing the multi-dimensional evaluation results by using a natural language processing model to generate structured evaluation text; Determining the creation target type according to the style evaluation result in the structured evaluation text; Obtaining recommended works from a preset reference work library by using a collaborative filtering algorithm according to the creation target type to generate personalized recommendation information.

7. The method according to claim 6, characterized in that, The processing the multi-dimensional evaluation results by using a natural language processing model to generate structured evaluation text includes: Extracting key evaluation features in the multi-dimensional evaluation results by using the semantic parsing layer of the natural language processing model, where the key evaluation features are determined based on the creation target type; Generating structured evaluation text including technical dimension evaluation descriptions, style dimension evaluations, and corresponding improvement suggestions by using the text generation layer of the natural language processing model based on the key evaluation features and a preset evaluation template.

8. An intelligent evaluation and feedback system for artistic creation, characterized in that, The device includes: A data acquisition module for obtaining the artistic work data uploaded by the user; A multi-modal feature analysis module for extracting multi-modal feature information including visual perception features and semantic features from the art work data; An intelligent evaluation and decision-making module for generating multi-dimensional evaluation results using a cross-modal attention mechanism based on the multi-modal feature information; A feedback generation and recommendation module for generating feedback information based on the multi-dimensional evaluation results.

9. A computer device, comprising a memory and a processor, the memory storing a computer program, characterized in that, When the processor executes the computer program, the steps of the method according to any one of claims 1 to 7 are implemented.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, the steps of the method according to any one of claims 1 to 7 are implemented.

Images