Method and terminal for analyzing a work of art

Abstract

The application discloses an art work analysis method and a terminal. Art work data input by a user is received and preprocessed, and a convolutional neural network is used to extract multi-level art features from the preprocessed art work data, so that multiple art key features of the art work, such as composition features, color features and brush stroke features, can be effectively extracted. Furthermore, multi-dimensional analysis results are generated according to the art key features, the analysis of the art work can be performed according to user requirements, and intelligent and interactive analysis of the art work is realized.

Description

Technical Field

[0001] This invention relates to the technical field of image analysis, and in particular to a method and terminal for analyzing works of art. Background Technology

[0002] Currently, artwork analysis in art education typically relies on teachers' verbal explanations and static image displays. Some educational institutions have begun using interactive whiteboards or projectors to display artwork, but these methods offer limited interactivity and intelligence.

[0003] Traditional methods of analyzing artworks lack interactivity and fail to engage students. Static displays cannot fully showcase the details and multi-dimensional information of the artworks, and teachers cannot provide personalized explanations based on the different levels of understanding of each student. Summary of the Invention

[0004] The technical problem to be solved by the present invention is to provide a method and terminal for analyzing works of art, which can analyze works of art according to user needs and realize intelligent and interactive analysis of works of art.

[0005] To solve the above-mentioned technical problems, the technical solution adopted by the present invention is as follows:

[0006] A method for analyzing a work of art, including the following steps:

[0007] S1. Receive artwork data input by the user and preprocess the artwork data;

[0008] S2. Use a convolutional neural network to extract multi-level art features from the pre-processed artwork data, and obtain multiple key art features by fusing and classifying the multi-level art features. The key art features include composition features, color features, and brushstroke features.

[0009] S3. Generate multidimensional analysis results of the artwork data based on multiple key artistic features.

[0010] To solve the above-mentioned technical problems, another technical solution adopted by the present invention is as follows:

[0011] An analysis terminal for an artwork includes a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, it implements the various steps of the aforementioned method for analyzing an artwork.

[0012] The beneficial effects of this invention are as follows: it receives and preprocesses the artwork data input by the user, and uses a convolutional neural network to extract multi-level art features from the preprocessed artwork data, thereby effectively extracting multiple key art features of the artwork, such as composition features, color features, and brushstroke features; and then generates multi-dimensional analysis results based on the key art features, enabling the analysis of artworks according to user needs, and realizing intelligent and interactive analysis of artworks. Attached Figure Description

[0013] Figure 1 A flowchart illustrating a method for analyzing artworks according to an embodiment of the present invention;

[0014] Figure 2 This is a schematic diagram of an analysis terminal for an artwork according to an embodiment of the present invention;

[0015] Figure 3 This is a block diagram illustrating a method for analyzing artworks according to an embodiment of the present invention;

[0016] Figure 4 This is a schematic diagram illustrating the overall steps of an art analysis method according to an embodiment of the present invention.

[0017] Label Explanation:

[0018] 1. An analysis terminal for artworks; 2. A memory; 3. A processor. Detailed Implementation

[0019] To explain in detail the technical content, objectives, and effects of the present invention, the following description is provided in conjunction with the embodiments and accompanying drawings.

[0020] Please refer to Figure 1 This invention provides a method for analyzing works of art, including the following steps:

[0021] S1. Receive artwork data input by the user and preprocess the artwork data;

[0022] S2. Use a convolutional neural network to extract multi-level art features from the pre-processed artwork data, and obtain multiple key art features by fusing and classifying the multi-level art features. The key art features include composition features, color features, and brushstroke features.

[0023] S3. Generate multidimensional analysis results of the artwork data based on multiple key artistic features.

[0024] As can be seen from the above description, the beneficial effects of the present invention are as follows: receiving and preprocessing the artwork data input by the user, using a convolutional neural network to extract multi-level art features from the preprocessed artwork data, thereby effectively extracting multiple key art features of the artwork; and then generating multi-dimensional analysis results based on the key art features, enabling the analysis of artworks according to user needs, and realizing intelligent and interactive analysis of artworks.

[0025] Further, step S2 includes:

[0026] Obtain a convolutional neural network model pre-trained using an image dataset, and use the convolutional neural network model to extract features at multiple levels respectively;

[0027] Multiple levels of features are fused, and the dimensionality of the fusion result is reduced using the principal component matrix. The dimensionality-reduced result is then clustered to obtain multiple key artistic features, which are then mapped to a 3D space for display.

[0028] As described above, convolutional neural network models can extract the characteristics of shallow, medium and deep layers respectively, and fuse, reduce the dimensionality and cluster the features of different layers. This can effectively extract and analyze the key features of artworks, including elements such as composition, color and brushstrokes.

[0029] Further, step S3 includes:

[0030] Each key feature of art is matched with preset aesthetic rules and art history databases to obtain matching results for each key feature of art. The matching results of each key feature of art are then integrated to obtain multidimensional analysis results.

[0031] As described above, by matching the analyzed key artistic features with preset rules and databases, more reasonable and accurate analysis results can be obtained, thus improving the reliability of the analysis.

[0032] Furthermore, step S3 also includes:

[0033] Receive the feature to be analyzed. If the key art features include the feature to be analyzed, then display the matching result corresponding to the feature to be analyzed.

[0034] As described above, based on the features to be analyzed provided by the user, the system can display the analysis results that the user needs, thereby providing personalized analysis for different users and improving the interactivity of the analysis.

[0035] Furthermore, it also includes:

[0036] S4. Construct a user preference vector based on the obtained user learning interaction operations, calculate the similarity between the user preference vector and the feature vector of the artwork in the artwork database, and recommend artworks to the user based on the similarity between the artwork and the user.

[0037] As described above, by using users' learning and interaction behaviors, highly similar artworks can be recommended to users in a targeted manner, thereby enabling dynamic and personalized artwork recommendations.

[0038] Please refer to Figure 2 Another embodiment of the present invention provides an analysis terminal for artworks, including a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, it implements the various steps of the above-described method for analyzing artworks.

[0039] The art analysis method and terminal described above are applicable to analyzing art works according to user needs, realizing intelligent and interactive analysis of art works. The following is a detailed description of specific implementation methods:

[0040] Please refer to Figure 1 , Figure 3 and Figure 4 Embodiment 1 of the present invention is as follows:

[0041] A method for analyzing a work of art, including the following steps:

[0042] S1. Receive artwork data input by the user and preprocess the artwork data.

[0043] Specifically, the artwork import module receives user-inputted artwork data. The image processing module adjusts the input artwork data to a uniform image size, for example, all images are adjusted to 224x224 pixels; then, the adjusted image is normalized, scaling the pixel values ​​to the range [0,1].

[0044] S2. Use a convolutional neural network to extract multi-level art features from the pre-processed artwork data, and obtain multiple key art features by fusing and classifying the multi-level art features. The key art features include composition features, color features, and brushstroke features.

[0045] Step S2 specifically involves: obtaining a convolutional neural network model pre-trained using an image dataset; extracting multiple hierarchical features using the convolutional neural network model; fusing the multiple hierarchical features; reducing the dimensionality of the fusion result using a principal component matrix; clustering the dimensionality reduction result to obtain multiple key artistic features; and mapping the multiple key artistic features to a 3D space for display.

[0046] Specifically, in this embodiment, a model pre-trained on a large-scale image dataset (such as ImageNet), such as VGG16 or ResNet50, is used as a feature extractor through the feature extraction module. Features are extracted from different layers of the CNN to capture information at different levels: shallow features such as edges and textures (F1); mid-level features such as shape and local structure (F2); and deep features at the semantic level (F3).

[0047] Features from different layers are concatenated and fused, and principal component analysis (PCA) is used to reduce the dimensionality of the fused feature matrix to reduce computational complexity. Then, the K-means algorithm is used to cluster the dimensionality-reduced features to identify key features. The t-SNE (t-distributed random neighbor embedding) algorithm can also be used to map high-dimensional features to 2D or 3D space for visualization.

[0048] By following the steps above, we can effectively extract and analyze the key features of artworks, including key elements such as composition, color, and brushstrokes.

[0049] S3. Generate multidimensional analysis results of the artwork data based on multiple key artistic features.

[0050] Step S3 specifically involves: matching each key art feature with preset aesthetic rules and art history databases through the artwork analysis module to obtain matching results corresponding to each key art feature, and integrating the matching results of each key art feature to obtain multidimensional analysis results.

[0051] Furthermore, the analysis results can be presented visually on large-screen devices through the interactive display module, and multiple interaction methods such as gestures and voice are provided.

[0052] In step S3, the feature to be analyzed can also be received. If the key art features include the feature to be analyzed, the matching result corresponding to the feature to be analyzed is displayed.

[0053] S4. Construct a user preference vector based on the obtained user learning interaction operations, calculate the similarity between the user preference vector and the feature vector of the artwork in the artwork database, and recommend artworks to the user based on the similarity between the artwork and the user.

[0054] Specifically, in order to make art recommendations based on user behavior and preferences, this embodiment adopts a hybrid recommendation method that combines collaborative filtering and content-based recommendation:

[0055] S41. User behavior data collection: Collect user browsing history, clicks, favorites, ratings, and other behavioral data.

[0056] S42. User preference modeling: Construct user preference vectors based on collected behavioral data.

[0057] The user preference vector P(u) is calculated as follows: P(u)=Σ(w_i*f_i) / Σw_i, where f_i represents the i-th behavioral feature and w_i represents the weight corresponding to the behavioral feature.

[0058] S43. Construction of feature vectors for artworks in the artwork database.

[0059] Specifically, constructing feature vectors requires considering multiple dimensions of the artwork, including visual features, art theory concepts, and emotional expression. The following are the steps and methods for constructing feature vectors:

[0060] S431. Basic visual feature extraction: Use pre-trained CNNs (such as VGG16 or ResNet) to extract low-level visual features; extract traditional visual features such as color histograms and edge orientation histograms.

[0061] S432. Art Theory Feature Calculation: Calculate the characteristic importance of the golden section point position, analyze the balance and symmetry of the picture; calculate the degree of color harmony based on color theory, analyze the contrast between warm and cool colors; analyze the length, direction and density distribution of brushstrokes.

[0062] S433. Semantic content analysis: Use object detection models to identify the main elements in the painting; classify scenes (such as landscapes, portraits, still lifes, etc.); when figures are present in the work, analyze their postures and expressions.

[0063] S434. Style Feature Extraction: Extract style features using the encoder of the style transfer network; calculate texture statistical features (such as the Gram matrix).

[0064] S435. Emotional Feature Analysis: Based on color psychology, analyze the emotions that a painting may convey; calculate the visual tension of the composition.

[0065] S436, Art Genre Classification: Obtain the probability distribution of art genres using a pre-trained art genre classification model.

[0066] S437. Technique Identification: Analyze whether a specific painting technique (such as pointillism, line drawing, etc.) is used; infer the painting materials used.

[0067] S438. Abstract quantification: Calculate the edge complexity of the image; analyze the richness of image details.

[0068] Concatenate all the above features into a high-dimensional vector; perform dimensionality reduction using PCA or t-SNE to obtain the final feature vector; standardize the final feature vector, which may look like the following:

[0069] [Visual Feature 1,..., Visual Feature n, Composition Feature 1,..., Composition Feature m, Color Feature 1,..., Color Feature k, Semantic Feature 1,..., Semantic Feature p, Style Feature 1,..., Style Feature q, Emotional Feature 1, Emotional Feature 2, School Probability 1,..., School Probability r, Technique Feature 1,..., Technique Feature s, Abstraction Level 1, Abstraction Level 2]

[0070] This feature vector not only contains objective visual information, but also incorporates art theory and subjective feelings, enabling a more comprehensive description of the characteristics of a work of art.

[0071] S44. Use user-based collaborative filtering to find similar users.

[0072] User similarity is calculated using cosine similarity:

[0073] sim(u,v)=P(u)·P(v) / (||P(u)||*||P(v)||)

[0074] In the formula, P(u) and P(v) are the preference vectors of users u and v, respectively.

[0075] S45. Content-based recommendation: Calculate the similarity between user preference vectors and work feature vectors.

[0076] Similarity calculation formula: sim(u,i)=P(u)·F(i) / (||P(u)||*||F(i)||);

[0077] In the formula, P(u) is the preference vector of user u, and F(i) is the feature vector of work i.

[0078] S46. Weighted fusion of the results from collaborative filtering and content-based recommendation.

[0079] Recommended score calculation formula: score(u,i)=α*CF(u,i)+(1-α)*CB(u,i)

[0080] In the formula, CF(u,i) is the collaborative filtering recommendation score, CB(u,i) is the content-based recommendation score, and α is the weight parameter.

[0081] S47. Sort the works according to the recommendation scores and generate the final recommendation list.

[0082] S48. Use online learning algorithms (such as online gradient descent) to update the user preference model in real time.

[0083] The updated formula is:

[0084] In the formula, η is the learning rate. It is the gradient of the loss function.

[0085] Through the above steps, artwork recommendations can be made dynamically and personally, and the recommendation results can be continuously adjusted and optimized based on user behavior and preferences.

[0086] Finally, the learning assessment module can track the user's learning process and generate learning reports and improvement suggestions.

[0087] The following provides a specific application scenario for analyzing artworks:

[0088] In an Impressionist painting appreciation class, the teacher used the terminal corresponding to this method for instruction. First, the teacher gave the voice command: "Show Monet's 'Impression, Sunrise'," and the method immediately displayed a high-definition image on the large screen. The teacher then said: "Please analyze the color characteristics of this painting," and the method automatically recognized the keyword "color," marked the main color blocks on the screen, and simultaneously displayed a color distribution pie chart next to them.

[0089] The teacher continued, "How did Monet depict light in this painting?" The method in this embodiment quickly marked the brushstrokes and color variations representing light on the canvas and displayed relevant technique explanations alongside them. A student zoomed in on a specific area of ​​the painting using a gesture; the method in this embodiment automatically identified the brushstroke characteristics of that area and provided a detailed explanation.

[0090] Another student asked, "How is this painting different from traditional landscape paintings?" This example immediately retrieved a traditional landscape painting for comparison, listing the differences in composition, color, and brushstrokes. After further explanation, the teacher asked the students to try creating a small piece using Impressionist techniques. The students created their works on a virtual canvas provided on the terminal, and their work was analyzed in real time, with suggestions for improvement given.

[0091] At the end of the course, the method of this embodiment generates a learning report for each student, including dimensions such as participation, comprehension, and creative level, and recommends suitable artworks and related materials for further study for each student.

[0092] Please refer to Figure 2 Embodiment two of the present invention is as follows:

[0093] An art analysis terminal 1 includes a memory 2, a processor 3, and a computer program stored on the memory 2 and executable on the processor 3. When the processor 3 executes the computer program, it implements the various steps of the art analysis method of Embodiment 1.

[0094] In summary, the present invention provides a method and terminal for analyzing artworks, which receives artwork data input by the user and preprocesses it, uses a convolutional neural network to extract multi-level art features from the preprocessed artwork data, thereby effectively extracting multiple key art features of the artwork; and then generates multi-dimensional analysis results based on the key art features, enabling the analysis of artworks according to user needs, and realizing intelligent and interactive analysis of artworks.

[0095] This invention enables students to interact with artworks directly in multiple dimensions through a large-screen intelligent assistant, significantly enhancing learning interest and engagement. It intelligently identifies and annotates key features of artworks, making abstract artistic concepts intuitive and visible, aiding students in better understanding and mastery. AI-based analysis and recommendation functions provide personalized learning content and suggestions for different students, improving learning efficiency. This approach provides teachers with powerful tools, allowing them to focus more on guiding and inspiring students, thus improving teaching quality. Furthermore, this invention cultivates students' art appreciation abilities and innovative thinking through comparative analysis and creative exercises. Automatically generated learning reports provide education administrators with objective and comprehensive evaluation data, contributing to continuous improvement of teaching methods and curriculum design.

[0096] The above description is merely an embodiment of the present invention and does not limit the patent scope of the present invention. Any equivalent modifications made based on the content of the present invention specification and drawings, or direct or indirect applications in related technical fields, are similarly included within the patent protection scope of the present invention.

Claims

1. A method for analyzing works of art, characterized in that, Including the following steps: S1. Receive artwork data input by the user and preprocess the artwork data; S2. Use a convolutional neural network to extract multi-level art features from the pre-processed artwork data, and obtain multiple key art features by fusing and classifying the multi-level art features. The key art features include composition features, color features, and brushstroke features. S3. Generate multidimensional analysis results of the artwork data based on multiple key artistic features.

2. The method for analyzing works of art according to claim 1, characterized in that, Step S2 includes: Obtain a convolutional neural network model pre-trained using an image dataset, and use the convolutional neural network model to extract features at multiple levels respectively; Multiple levels of features are fused, and the dimensionality of the fusion result is reduced using the principal component matrix. The dimensionality-reduced result is then clustered to obtain multiple key artistic features, which are then mapped to a 3D space for display.

3. The method for analyzing works of art according to claim 1, characterized in that, Step S3 includes: Each key feature of art is matched with preset aesthetic rules and art history databases to obtain matching results for each key feature of art. The matching results of each key feature of art are then integrated to obtain multidimensional analysis results.

4. The method for analyzing a work of art according to claim 3, characterized in that, Step S3 also includes: Receive the feature to be analyzed. If the key art features include the feature to be analyzed, then display the matching result corresponding to the feature to be analyzed.

5. The method for analyzing works of art according to claim 1, characterized in that, Also includes: S4. Construct a user preference vector based on the obtained user learning interaction operations, calculate the similarity between the user preference vector and the feature vector of the artwork in the artwork database, and recommend artworks to the user based on the similarity between the artwork and the user.

6. A terminal for analyzing works of art, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the computer program, it performs the following steps: S1. Receive artwork data input by the user and preprocess the artwork data; S2. Use a convolutional neural network to extract multi-level art features from the pre-processed artwork data, and obtain multiple key art features by fusing and classifying the multi-level art features. The key art features include composition features, color features, and brushstroke features. S3. Generate multidimensional analysis results of the artwork data based on multiple key artistic features.

7. The art analysis terminal according to claim 6, characterized in that, Step S2 includes: Obtain a convolutional neural network model pre-trained using an image dataset, and use the convolutional neural network model to extract features at multiple levels respectively; Multiple levels of features are fused, and the dimensionality of the fusion result is reduced using the principal component matrix. The dimensionality-reduced result is then clustered to obtain multiple key artistic features, which are then mapped to a 3D space for display.

8. The art analysis terminal according to claim 6, characterized in that, Step S3 includes: Each key feature of art is matched with preset aesthetic rules and art history databases to obtain matching results for each key feature of art. The matching results of each key feature of art are then integrated to obtain multidimensional analysis results.

9. The art analysis terminal according to claim 8, characterized in that, Step S3 also includes: Receive the feature to be analyzed. If the key art features include the feature to be analyzed, then display the matching result corresponding to the feature to be analyzed.

10. The art analysis terminal according to claim 6, characterized in that, Also includes: S4. Construct a user preference vector based on the obtained user learning interaction operations, calculate the similarity between the user preference vector and the feature vector of the artwork in the artwork database, and recommend artworks to the user based on the similarity between the artwork and the user.

Images