A painting color association analysis method and system based on bipartite network

Abstract

The invention discloses a method and system for analyzing the association of painting colors based on a bipartite network, including constructing a color analysis system based on a bipartite network. The method and system for analyzing the association of painting colors based on a bipartite network realize efficient and accurate color analysis of Chinese paintings by constructing a color association model based on a bipartite network, combining the CIELAB2000 color difference standard and a rich interactive visualization tool. By directly focusing on color nodes through a network projection method, the analysis efficiency of color relationships is effectively improved, and the workload of researchers in image processing and color relationship discovery is reduced. The method can accurately reflect the differences in brightness, chromaticity and hue of colors, and ensure the scientificity and reliability of the analysis results. At the same time, the integrated visualization system enables researchers to intuitively explore and verify color relationships from multiple dimensions, and is widely applicable to color research in the field of digital humanities, especially in the color analysis of traditional Chinese paintings. It has important application value.

Description

Technical Field

[0001] The present invention relates to the technical field of digital humanities and big data visualization analysis, and specifically to a painting color association analysis method and system based on a bipartite network. Background Art

[0002] The color culture of traditional Chinese painting is an important legacy of contemporary art, and its color composition involves multiple factors of traditional Chinese aesthetics, philosophy, culture, and artistic concepts. Color analysis can not only reveal the historical and cultural connotations of painting, but also provide important clues for understanding the artistic development of painting. However, existing color analysis methods mainly focus on artistic or aesthetic interpretations, and often rely on qualitative analysis of color groups and their associations. Although this method can help understand the use of color combinations in specific scenarios, its limitation is that it cannot effectively deal with the complex color patterns in traditional Chinese paintings;

[0003] Prior art 1 (application number CN201711224874.5, Chinese patent application date 2017-11-29) is an image feature retrieval method based on similarity association analysis, which vectorizes and regionalizes image objects in a coordinate system, performs location analysis on the boundaries and shared points of the segmented regions, and then performs coordinate system location and image color analysis on the intersection image part, and through orderly image region interference location analysis and color feature analysis, image objects with high similarity are retrieved more efficiently and accurately;

[0004] There is also prior art 2 (application number CN201610645563.5, Chinese patent application date 2016-08-05) in which a detection and evaluation method for measuring the damage of lighting sources to the color of traditional paintings is provided. According to the decay of the color parameter values ​​of traditional Chinese paintings, an analysis and evaluation method for quantitatively evaluating the degree of influence of different light sources on the color of paintings is provided, and the color decay law of different types of paintings under the illumination of various light sources is obtained, and the quantitative influence degree and relative influence coefficient of various light sources on different types of paintings are obtained;

[0005] And prior art 3 (Chinese patent with application number CN201910195692.2 and application date 2019-03-15) is a method for visually expressing the color characteristics of Chinese paintings, which effectively solves the problem of non-intuitive display of color attributes; second, the quantitative data display of the color characteristics of Chinese paintings provides designers of Chinese-style cultural and creative products with multi-dimensional reference information on color schemes and color ratio perception, effectively reducing their time spent on color matching and effectively solving the problem of time-consuming color matching selection; third, the addition of interactive functions of the present invention can allow users with different needs to independently choose the information they want to see;

[0006] Most current color network analysis is based on clustering methods, which are suitable for color-dense areas. However, its effect is limited in Chinese paintings with rich color variations. In addition, existing methods usually rely on the frequency of color occurrence to build networks, which may cause some rare but important colors to be ignored. Researchers usually need to constantly compare, search and verify the relationship between colors in a large number of images. This process is not only time-consuming and laborious, but also easy to miss important color associations. Therefore, there is an urgent need for a comprehensive analysis model that can combine color and painting content to help researchers efficiently discover and analyze complex color relationships.

[0007] Therefore, we proposed a painting color association analysis method and system based on bipartite network to solve the above problems. Summary of the invention

[0008] The purpose of the present invention is to provide a method and system for analyzing the color association of paintings based on a bipartite network, so as to solve the problem raised by the above-mentioned background technology that the existing analysis technology in the market is difficult to explore the association between paintings and colors, and between colors. The present invention provides a method and system for visual analysis based on a bipartite network, which refines, associates and filters paintings and the colors therein through a network structure, thereby clarifying the complex color transformation rules.

[0009] To achieve the above object, the present invention provides the following technical solution: a method and system for analyzing the association of painting colors based on a bipartite network, comprising the following steps:

[0010] (1) Constructing a color analysis system based on a bipartite network, the system is connected to a color database, and the source images of paintings in the database are converted into LAB color space to support subsequent processing. The color analysis system has an embedded color extraction and association calculation module;

[0011] (2) The color extraction module is used to extract the color ratio distribution of each painting. First, the color difference is calculated based on the CIELAB2000 color difference standard and the distribution ratio of the painting colors is obtained through clustering. Then, the similarity between the colors of the paintings is calculated to merge the colors and then construct a color binary network.

[0012] (3) The association calculation module is used to calculate the color association between different paintings. Based on the bipartite network of painting and color dependency, the bipartite network is transformed into a single-mode network through a projection algorithm, and the painting dimensions are compressed to generate the association between colors. The importance of different colors in the network is calculated through network centrality.

[0013] (4) Based on the color network and related color information obtained in step (3), a visualization view and filtering query interaction component are designed for the color analysis system to help researchers intuitively explore and analyze color associations.

[0014] Preferably, the LAB color space conversion of the painting source image in step (1) specifically includes the following steps:

[0015] Convert the RGB color space into the LAB color space. First, convert the RGB color space into the CIE1931XYZ color space as a transition state.

[0016] Normalize RGB, standardize RGB values ​​to the range of [0, 1];

[0017] Gamma correction is performed on the normalized RGB values ​​to offset the nonlinear effects of the display;

[0018] Use the standard RGB to XYZ conversion matrix (based on the reference white point D65 of the sRGB color space) to complete the linear RGB to XYZ color space conversion;

[0019] The XYZ color space is converted into the CIELAB color space, and the conversion is completed based on the conversion formula based on D65 (under standard light source) as the reference white point of the CIELAB color space to obtain the L, a, and b values ​​obtained by converting R, G, and B.

[0020] Preferably, the color extraction module in step (2) specifically includes the following steps:

[0021] For color extraction of a painting, the color difference between any two color pixels is calculated using the CIELAB2000 standard, and a discrimination threshold is specified;

[0022] If the color difference between two pixels is less than the specified threshold, the clusters are considered to be the same color;

[0023] If the color difference between two pixels is greater than the specified threshold, they are not clustered;

[0024] After traversing all the pixels, we get the representative color and its proportion of a painting after clustering. The representative color is the color with the largest proportion in a group of colors whose color difference is less than the threshold. The proportion of the representative color is the proportion of all color pixels in the group to all pixels in the painting.

[0025] For the potential color correlation between any two paintings, the color difference between any colors in the two paintings is calculated using the CIELAB2000 standard, and a discrimination threshold is specified;

[0026] If the color difference between two colors is less than the specified threshold, the two colors are merged into the same color, and both paintings are associated with the color;

[0027] If the color difference between two colors is greater than the specified threshold, they will not be merged;

[0028] Abstract all paintings into painting nodes, and abstract all colors into color nodes; construct a painting and color binary network based on the relationship between paintings and colors.

[0029] Preferably, the association calculation module in step (3) specifically includes the following steps:

[0030] The bipartite network of paintings and colors is transformed into a color unimodal network through the projection algorithm of the bipartite network; the color unimodal network is composed entirely of color nodes, which optimizes the dimension of paintings; the edge weights between colors are determined by the frequency of each color appearing in all paintings and the frequency of two colors appearing in the same painting;

[0031] The more frequently two colors appear in the same painting, and the lower the frequency of each color appearing in all paintings, the higher the edge weight calculated in the end.

[0032] Preferably, the bipartite network is transformed into a color-centered single-mode network based on the dependency between paintings and colors and the similarity between colors; the importance and evolution relationship of colors in the color single-mode network are calculated by network centrality, including:

[0033] Calculate the number of connections between a color node and other color nodes to measure degree centrality;

[0034] Calculate the closeness of a color node to other color nodes to measure closeness centrality.

[0035] Preferably, the color visual analysis system and its view in step (4) specifically include the following steps:

[0036] The color visual analysis system supports the selection of paintings based on subject matter, material, and shape keywords and conducts subsequent painting color analysis. The visualization interface is built based on the Echarts library. Its specific views and functions are as follows: network view, color space view, color distribution view, and monochrome query view.

[0037] Preferably, network view: the network view presents all paintings that meet the filtering conditions and the color distribution information in the paintings, and visualizes the corresponding paintings and color nodes in the form of a bipartite network. The view supports selecting color differences within a specified range to further merge the colors between paintings or selecting any edge weights within a specified range to guide the projection algorithm to transform the painting and color bipartite network into a color single-mode network. The network view also provides two quantitative indicators, height centrality and close centrality, to show the degree of connection and importance between different colors in all color nodes under the filtering conditions.

[0038] Preferably, color space view: The color space view takes RGB and LAB color spaces as the basis, presents each color node and its proportion in the form of 3D scattered points, and supports verification of the specific source of each color.

[0039] Preferably, color distribution view: the color distribution view is used to statistically analyze all color nodes under filtering conditions and the subject matter, dynasty and color setting information of the paintings to which they are attached, and is used to illustrate the color usage characteristics and differences of paintings of different dynasties and themes.

[0040] Preferably, monochrome query view: The monochrome query view supports users to conduct in-depth mining on the monochrome of interest, by specifying the RGB value of a single color in the form of a tree diagram to derive potential similar colors, and then select a specific color to query all paintings involved in the color and related subject matter and dynasty information.

[0041] Compared with the prior art, the beneficial effects of the present invention are as follows: by constructing a color association model based on a bipartite network, combined with the CIELAB2000 color difference standard and a rich interactive visualization tool, efficient and accurate color analysis of Chinese paintings is achieved. By directly focusing on color nodes through a network projection method, the analysis efficiency of color relationships is effectively improved, and the workload of researchers in image processing and color relationship discovery is reduced. The method can accurately reflect the brightness, chromaticity and hue differences of colors, ensuring the scientificity and reliability of the analysis results. At the same time, the integrated visualization system enables researchers to intuitively explore and verify color relationships from multiple dimensions, and is widely applicable to color research in the field of digital humanities, especially in the color analysis of traditional Chinese paintings. It has important application value. BRIEF DESCRIPTION OF THE DRAWINGS

[0042] Figure 1 Analyze the flow chart for the present invention;

[0043] Figure 2 It is a schematic diagram of the painting, the color binary network and the color single mode network of the present invention;

[0044] Figure 3 It is a schematic diagram of the components of the visualization system of the present invention;

[0045] Figure 4 Schematic diagram of the relationship between paintings and colors mined by the present invention. DETAILED DESCRIPTION

[0046] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. 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 creative work are within the scope of protection of the present invention.

[0047] The present invention provides the following technical solutions:

[0048] The steps include:

[0049] (1) Constructing a color analysis system based on a bipartite network, the system is connected to a color database, and the source images of paintings in the database are converted into LAB color space to support subsequent processing. The color analysis system has an embedded color extraction and association calculation module;

[0050] (2) The color extraction module is used to extract the color ratio distribution of each painting. First, the color difference is calculated based on the CIELAB2000 color difference standard and the distribution ratio of the painting colors is obtained through clustering. Then, the similarity between the colors of the paintings is calculated to merge the colors and then construct a color binary network.

[0051] (3) The association calculation module is used to calculate the color association between different paintings. Based on the bipartite network of painting and color dependency, the bipartite network is transformed into a single-mode network through a projection algorithm, and the painting dimensions are compressed to generate the association between colors. The importance of different colors in the network is calculated through network centrality.

[0052] (4) Based on the color network and related color information obtained in step (3), a visualization view and filtering query interaction component are designed for the color analysis system to help researchers intuitively explore and analyze color associations.

[0053] The LAB color space conversion of the painting source image in step (1) specifically includes the following steps:

[0054] Convert the RGB color space into the LAB color space. First, convert the RGB color space into the CIE1931XYZ color space as a transition state.

[0055] Normalize RGB, standardize RGB values ​​to the range of [0, 1];

[0056] Gamma correction is performed on the normalized RGB values ​​to offset the nonlinear effects of the display;

[0057] Use the standard RGB to XYZ conversion matrix (based on the reference white point D65 of the sRGB color space) to complete the linear RGB to XYZ color space conversion;

[0058] The XYZ color space is converted into the CIELAB color space, and the conversion is completed based on the conversion formula based on D65 (under standard light source) as the reference white point of the CIELAB color space to obtain the L, a, and b values ​​obtained by converting R, G, and B.

[0059] The color extraction module in step (2) specifically includes the following steps:

[0060] For color extraction of a painting, the color difference between any two color pixels is calculated using the CIELAB2000 standard, and a discrimination threshold is specified;

[0061] If the color difference between two pixels is less than the specified threshold, the clusters are considered to be the same color;

[0062] If the color difference between two pixels is greater than the specified threshold, they are not clustered;

[0063] After traversing all the pixels, we get the representative color and its proportion of a painting after clustering. The representative color is the color with the largest proportion in a group of colors whose color difference is less than the threshold. The proportion of the representative color is the proportion of all color pixels in the group to all pixels in the painting.

[0064] For the potential color correlation between any two paintings, the color difference between any colors in the two paintings is calculated using the CIELAB2000 standard, and a discrimination threshold is specified;

[0065] If the color difference between two colors is less than the specified threshold, the two colors are merged into the same color, and both paintings are associated with the color;

[0066] If the color difference between two colors is greater than the specified threshold, they will not be merged;

[0067] Abstract all paintings into painting nodes, and abstract all colors into color nodes; construct a painting and color binary network based on the relationship between paintings and colors.

[0068] The association calculation module in step (3) specifically includes the following steps:

[0069] The bipartite network of paintings and colors is transformed into a color unimodal network through the projection algorithm of the bipartite network; the color unimodal network is composed entirely of color nodes, which optimizes the dimension of paintings; the edge weights between colors are determined by the frequency of each color appearing in all paintings and the frequency of two colors appearing in the same painting;

[0070] The more frequently two colors appear in the same painting, and the lower the frequency of each color appearing in all paintings, the higher the edge weight calculated in the end.

[0071] Based on the dependency between paintings and colors and the similarity between colors, the bipartite network is transformed into a color-centered single-mode network; the importance and evolution relationship of colors in the color single-mode network are calculated through network centrality, including:

[0072] Calculate the number of connections between a color node and other color nodes to measure degree centrality;

[0073] Calculate the closeness of a color node to other color nodes to measure closeness centrality.

[0074] The color visual analysis system and its view in step (4) specifically include the following steps:

[0075] The color visual analysis system supports the selection of paintings based on subject matter, material, and shape keywords and conducts subsequent painting color analysis. The visualization interface is built based on the Echarts library. Its specific views and functions are as follows: network view, color space view, color distribution view, and monochrome query view.

[0076] Network view: The network view presents all paintings that meet the filtering conditions and the color distribution information in the paintings, and visualizes the corresponding paintings and color nodes in the form of a bipartite network. The view supports selecting color differences within a specified range to further merge the colors between paintings, or selecting any edge weights within a specified range to guide the projection algorithm to transform the painting and color bipartite network into a color single-mode network. The network view also provides two quantitative indicators, height centrality and close centrality, to show the degree of connection and importance between different colors in all color nodes under the filtering conditions.

[0077] Color space view: The color space view uses RGB and LAB color spaces as the basis, presents each color node and its proportion in the form of 3D scattered points, and supports verification of the specific source of each color.

[0078] Color distribution view: Color distribution view is used to count all color nodes under the filtering conditions and the subject matter, dynasty and color setting information of the paintings to which they are attached, and to illustrate the color usage characteristics and differences of paintings of different dynasties and themes.

[0079] Monochrome query view: Monochrome query view allows users to conduct in-depth research on the single color of interest. By specifying the RGB value of a single color in the form of a tree diagram, users can derive potential similar colors. Then, users can select a specific color to query all paintings related to the color and related subject matter and dynasty information.

[0080] Embodiment 1: The source painting image RGB color space is converted, as shown in the attached Figure 1 As shown:

[0081] First, the painting source image in the database is converted into the LAB color space to support subsequent processing. Step (1) of the technical solution is implemented by the following sub-steps:

[0082] Convert the RGB color space in the painting into CIE1931XYZ color space;

[0083] Normalize RGB to normalize RGB values ​​to the range [0, 1]:

[0084]

[0085] Due to the nonlinear effect of the display, gamma correction is performed on the normalized RGB values

[0086]

[0087] Correct the G and B components in the same way

[0088] The linear RGB to XYZ conversion is done based on the standard RGB to XYZ conversion matrix (based on the reference white point D65 of the sRGB color space):

[0089]

[0090] Convert the CIE1931XYZ color space in the painting into the CIELAB color space;

[0091] Based on D65 (under standard light source) as the reference white point of CIELAB color space, the reference white point value of D65 is:

[0092] X n =95.047,Y n =100.000,Z n =108.883

[0093] Convert the XYZ color space to the CIELAB color space and calculate the L, a, and b values:

[0094]

[0095] Example 2: Extracting the color distribution of a painting image and constructing a painting-color binary network:

[0096] After converting the RGB color space into the LAB color space, the color difference between any two pixels in the source image of the painting is calculated using the CIELAB2000 standard.

[0097]

[0098] If the color difference between two pixels is less than the specified threshold, the clusters are considered to be the same color.

[0099] If the color difference between two pixels is greater than the specified threshold, they are not clustered.

[0100] After traversing all the pixels, the representative color and proportion of each color group are calculated based on all the color groups after clustering. The representative color is the color with the largest proportion in a group of colors whose color difference is less than the threshold, and the proportion of the representative color is the proportion of all color pixel points in the group to all pixel points in the painting.

[0101] In order to further strengthen the association between the paintings, we need to use the similarity between the color distributions of the paintings; specify the color threshold, and calculate the difference between all clustered colors of the two paintings. If there are two colors in the two paintings with a color difference less than the threshold, they are merged, and both paintings are associated with the merged color.

[0102] Figure 2 A total of 614 paintings from the Tang and Song dynasties were selected as cases for color extraction and binary network construction, where parts (a), (b), and (c) are binary networks for Tang and Song dynasty paintings with color difference thresholds of 3, 5, and 7, respectively.

[0103] All paintings are analyzed in this way to obtain the paintings and their corresponding color distribution, with paintings as one type of node in the network and colors as another type of node. The colors in each painting are linked to the corresponding color nodes according to their proportion in the image, forming an association relationship between paintings and colors, and completing the binary network between paintings and colors.

[0104] Example 3: Converting the painting and color binary network into a color single-mode network:

[0105] The bipartite network of paintings and colors is transformed into a color unimodal network through the projection algorithm of the bipartite network. The projection process uses the following formula to calculate the edge weight between two painting nodes:

[0106]

[0107] Where u and v are any two colors, m is the total number of paintings, and if color u does not exist in painting k, x uk =0, otherwise x uk = 1. Similarly, if color v does not exist in painting k, x vk =0, otherwise x vk =1.

[0108] Thus, the bipartite network is transformed into a color-centered single-mode network based on the dependency between paintings and colors and the similarity between colors. Figure 2 (d) shows the color single-mode network directly transformed from the binary network of Tang and Song Dynasty paintings with a color difference threshold of 3. The network structure can be further optimized by filtering the edge weights; Figure 2 (e)(f) show the single-mode color network after filtering the edge weights less than 0.5 and less than 0.9, respectively.

[0109] The importance and evolution relationship of colors in color unimodal networks need to be calculated through network centrality, including:

[0110] Degree centrality measures the number of connections between a color node and other color nodes

[0111]

[0112] Closeness centrality measures the closeness of a color node to other color nodes

[0113]

[0114] Embodiment 4, color visual analysis system design;

[0115] Design a color visual analysis system based on a bipartite network, which supports the selection of paintings by subject matter, material, and form keywords and conducts subsequent painting color analysis. On this basis, it provides multiple views for interaction. Figure 3 The relevant views in the color visual analysis system are jointly shown:

[0116] The network visualization view provides researchers with a visualization of color relationships. Through the color-painting bipartite network, researchers can intuitively see the distribution of colors in each painting and its association with other colors. Furthermore, through the projection of the single-mode color network, researchers can focus on the relationship between colors and adjust the network weights to reveal which colors have strong co-occurrence relationships in multiple paintings. With this feature, researchers can identify the most common color combinations to better understand the color matching logic and composition patterns in traditional Chinese paintings.

[0117] The color node attribute view provides detailed information about each color node, and researchers can obtain key indicators of specific colors in the network through components, such as degree centrality and closeness centrality. For example, by analyzing degree centrality and betweenness centrality, we can reveal the bridge role of certain colors between different paintings, helping researchers better understand how these colors are reused in multiple painting forms across different periods and styles.

[0118] The color space view provides researchers with a global view of all colors. The view shows all the colors extracted from the database, and researchers can interactively click on any color to view its RGB / LAB value and the corresponding painting. This not only allows researchers to view the precise technical information of the color, but also to relate this information to the content of the actual work, laying the foundation for further analysis. The color space view also provides researchers with the most original color data, which can serve as the core foundation for subsequent color query and analysis.

[0119] The color attribute distribution view further expands the functionality of color analysis. The view uses a Sankey diagram to show the path flow between color and real-world attributes, revealing the association between color and painting themes, similar colors, and painting method attributes. This feature is particularly suitable for studying the use of color in a specific cultural or artistic context. For example, by tracking the historical use frequency and painting themes of certain colors, you can gain a deeper understanding of the cultural symbols and artistic significance of these colors in different historical periods.

[0120] The color query component provides researchers with powerful monochrome query and painting query functions. Researchers can find related colors by entering RGB values ​​and generate a hierarchy to reveal the similarity relationship between colors. In addition, the painting query function allows researchers to deeply explore the use of a certain color in a specific painting, which is of great significance for studying the color composition of a specific painting or tracing the historical development path of a certain color. Researchers can further view which paintings the color node appears in, view the detailed information of the painting through the slider function, and further explore the application of color in a specific painting.

[0121] Taking the analysis of the performance of cyan in Song Dynasty paintings as an example, we first selected "Song Dynasty" as the research object through the scope determination component, and used the system's color query function to locate colors similar to cyan in the database by entering specific RGB values ​​(69, 130, 96). This precise color query method allows researchers to quickly find the color node closest to the specified color and generate a detailed hierarchy showing similar colors at different levels. Through this hierarchy, the researchers found that the color nodes closest to the predefined cyan were widely distributed, but these colors were relatively dim in actual paintings and had no strictly consistent standard colors. This shows that in Song Dynasty paintings, even the same cyan may show subtle differences depending on the background, subject or pigment material of the painting. These findings not only reveal the diversity of cyan, but also suggest the flexibility of Song Dynasty painters in using the same hue. The researchers further used the color node attribute function to deeply analyze the importance of cyan in the color network. By looking at the centrality indicators of the nodes (such as degree centrality and closeness centrality), the researchers learned that the position of cyan in the network shows that it occupies a relatively core position in multiple paintings. This shows that cyan did not appear by chance in Song Dynasty paintings, but as a common and important color, it played a key role in different works, especially in landscape paintings and certain specific painting styles, where it was used more frequently. In order to further understand the practical application of cyan, the researchers used the system function to query which specific paintings these cyan nodes appeared in. After in-depth exploration, the system finally displayed four paintings containing cyan, such as Figure 4The researchers conducted a detailed comparative analysis of these paintings and found that cyan was used differently in different paintings. For example, in landscape paintings, cyan is often used to depict mountains and vegetation, conveying the tranquility and depth of nature; while in flower and bird paintings, cyan is more often used as a background color to set off other bright colors. These differences in color use may be closely related to the aesthetic ideas, cultural background and personal style of the painters in the Song Dynasty.

[0122] Although the present invention has been described in detail with reference to the aforementioned embodiments, it is still possible for those skilled in the art to modify the technical solutions described in the aforementioned embodiments, or replace some of the technical features therein. Any modification, replacement, and improvement made within the spirit and principles of the present invention should be included in the protection scope of the present invention.

Claims

1. A method for analyzing the association of painting colors based on a bipartite network, comprising the following steps: (1) constructing a color analysis system based on a bipartite network, the system is connected to a color database, and the source images of paintings in the database are converted into LAB color space to support subsequent processing, and the color analysis system has an embedded color extraction and association calculation module; (2) the color extraction module is used to extract the color ratio distribution of each painting, firstly calculate the color difference based on the CIELAB2000 color difference standard and obtain the distribution ratio of the painting colors through clustering, and then calculate the similarity between the painting colors to merge the colors, and then construct a color bipartite network; (3) The association calculation module is used to calculate the color association between different paintings. Based on the bipartite network of painting and color dependency, the bipartite network is transformed into a single-mode network through a projection algorithm, and the painting dimensions are compressed to generate the association between colors. The importance of different colors in the network is calculated through network centrality. (4) Based on the color network and related color information obtained in step (3), a visualization view and filtering query interactive component are designed for the color analysis system to help researchers intuitively explore and analyze color associations; the association calculation module in step (3) specifically includes the following steps: the bipartite network of paintings and colors is converted into a color single-mode network through a bipartite network projection algorithm; the color single-mode network is completely composed of color nodes, which optimizes the painting dimension; The edge weights between colors are determined by the frequency of each color appearing in all paintings and the frequency of the two colors appearing in the same painting. The higher the frequency of the two colors appearing in the same painting and the lower the frequency of the two colors appearing in all paintings, the higher the edge weight calculated. Based on the dependency between paintings and colors and the similarity between colors, the bipartite network is transformed into a color-centered single-mode network. The importance and evolution relationship of colors in the color single-mode network are calculated through network centrality, including: calculating the number of connections between a color node and other color nodes to measure degree centrality; Calculate the closeness of a color node to other color nodes to measure closeness centrality.

2. The method for analyzing the association of painting colors based on a bipartite network according to claim 1, characterized in that: The LAB color space conversion of the painting source image in step (1) specifically includes the following steps: converting the RGB color space into the LAB color space, first converting the RGB color space into the CIE1931XYZ color space as a transition state; Normalize RGB to standardize the RGB values ​​to the range of [0, 1]; perform gamma correction on the standardized RGB values ​​to offset the nonlinear effect of the display; Use the RGB to XYZ conversion matrix based on the reference white point D65 standard of the sRGB color space to complete the conversion from linear RGB to XYZ color space; convert the XYZ color space to CIELAB color space, and complete the conversion based on the conversion formula based on D65 as the reference white point of the CIELAB color space under standard light source to obtain the L, a, b values ​​obtained by the conversion of R, G, B.

3. The method for analyzing the association of painting colors based on a bipartite network according to claim 1, characterized in that: The color extraction module in step (2) specifically includes the following steps: for color extraction of a painting, the color difference between any two color pixels is calculated according to the CIELAB2000 standard, and a discrimination threshold is specified; if the color difference between the two pixels is less than the specified threshold, the two pixels are clustered and regarded as the same color; If the color difference between two pixels is greater than the specified threshold, they are not clustered; after traversing all the pixels, the representative color and its proportion of a painting after clustering are obtained. The representative color is the color with the largest proportion in a group of colors with a color difference less than the threshold, and the proportion of the representative color is the proportion of all color pixels in the group to all pixels in the painting; for the potential color association between any two paintings, the color difference between any colors in the two paintings is calculated using the CIELAB2000 standard, and a discrimination threshold is specified; if the color difference between two colors is less than the specified threshold, the two colors are merged into the same color, and both paintings are associated with the color; If the color difference between two colors is greater than the specified threshold, they will not be merged; all paintings are abstracted into painting nodes, and all colors are abstracted into color nodes; a painting and color binary network is constructed based on the association between paintings and colors.

4. A painting color association analysis system based on a bipartite network, characterized in that: A method for analyzing the association of painting colors based on a bipartite network as described in claim 1 is adopted, and the specific method steps of the system are as follows: The color visual analysis system and its view in step (4) specifically include the following steps: The color visual analysis system supports the selection of paintings based on keywords such as subject matter, material, and shape, and conducts subsequent painting color analysis. The visualization interface is built based on the Echarts library, and its specific views and functions are as follows: network view, color space view, color distribution view, and monochrome query view.

5. The painting color association analysis system based on bipartite network according to claim 4 is characterized by: The network view presents all paintings that meet the filtering conditions and the color distribution information in the paintings, and visualizes the corresponding paintings and color nodes in the form of a bipartite network. The view supports selecting color differences within a specified range to further merge the colors between paintings or selecting any edge weights within a specified range to guide the projection algorithm to transform the painting and color bipartite network into a color single-mode network. The network view also provides two quantitative indicators, height centrality and close centrality, to show the degree of connection and importance between different colors in all color nodes under the filtering conditions.

6. The painting color association analysis system based on bipartite network according to claim 4 is characterized by: The color space view uses RGB and LAB color spaces as the basis, presents each color node and its proportion in the form of 3D scatter points, and supports verification of the specific source of each color.

7. The painting color association analysis system based on bipartite network according to claim 4 is characterized by: The color distribution view is used to count all color nodes under the filtering conditions and the subject matter, dynasty and color setting information of the paintings to which they are attached, and to illustrate the color usage characteristics and differences of paintings of different dynasties and themes.

8. The painting color association analysis system based on bipartite network according to claim 4 is characterized by: The monochrome query view allows users to conduct in-depth research on monochrome colors of interest. By specifying the RGB value of a single color in the form of a tree diagram, users can derive potential similar colors. Then, by selecting a specific color, users can query all paintings related to the color and related subject matter and dynasty information.

Images