A method and system for distinguishing the authenticity of seals on calligraphy and paintings

Abstract

The present invention discloses a method and system for identifying the authenticity of seals on calligraphy and paintings. The method includes the steps of: S1, collecting the image information of the original work and establishing a corresponding database of the seal and the original work; S2, constructing a seal portrait from three dimensions of the overall characteristics, local characteristics and associated characteristics of the seal; S3, constructing the associated relationship between the seals in the work, and combining the collection seal with the existing work age information to obtain the total seal map, relationship map, clustering map and age timeline; S4, forming the comprehensive similarity index of the seal according to the weighted average of the three dimensions, and comprehensively judging the authenticity of the seal work. The present invention comprehensively judges the similarity index from multiple feature dimensions and work age information, and solves the problem in the prior art of lacking a comprehensive analysis of the similarity index of the seal from multiple dimensions of features.

Description

Technical Field

[0001] The present invention belongs to the technical field of image recognition. Specifically, it relates to a method and system for distinguishing the authenticity of seals on calligraphy and paintings. Background Art

[0002] The seals in ancient Chinese calligraphy and painting works contain a lot of important information, providing an important basis for the authenticity identification of calligraphy and painting works. Therefore, for humanists, the comparison and analysis of seals are very important. However, due to a series of problems existing in existing calligraphy and painting works and technical methods, the recognition and appreciation of seals face huge challenges. First, the number of existing artworks and their seals is extremely large, making it very time-consuming and laborious for experts to manually compare and analyze the seals. Second, existing calligraphy and painting works come from museums around the world, and access to the data is restricted by the regulations of each museum, and the data sources are very unstable. Third, currently, the seal images cannot be viewed separately and must be viewed in the original works. Finally, the existing data is relatively scattered, lacking a unified organizational form and retrieval system. In addition, there is no correlation between calligraphy and painting works and seals, and it is impossible to quickly and accurately screen out the information of calligraphy and painting works required for appreciation and recognition.

[0003] In a Chinese invention patent with the patent number CN201310120418.1, a calligraphy and painting work authenticity appraisal system and method are disclosed. The appraisal method includes: observing and analyzing the stable and regular habitual characteristics when the author created from genuine works; observing and analyzing the personal characteristics of the author when creating and the morphological characteristics in genuine works; observing and analyzing the writing habit characteristics of the author when creating, the text style characteristics and detail characteristics in genuine works; observing and analyzing the material characteristics of genuine works and the defect characteristics in genuine works; taking any one, several or all of the above characteristics as appraisal characteristics, photographing the images of the appraisal areas where the appraisal characteristics are located, and obtaining the image feature data of genuine works accordingly, and storing them to form an information file; comparing and analyzing the image features for appraisal in each of the appraisal areas of the work to be authenticated with the corresponding image features of the pre-stored genuine works in each of the appraisal areas, and using this as the basis for judging the authenticity of the work to be authenticated. Simplifying the complex image characteristics of calligraphy and painting works into systematic and concise image features or their data information that are easy to judge and express can perform authenticity appraisal simply and quickly.

[0004] The defect of the existing patent is that although it realizes the simple, fast and effective determination of the authenticity of calligraphy and painting works, it lacks a comprehensive analysis similarity index for the characteristics of seals in multiple dimensions. Summary of the Invention

[0005] In view of the problem that there is a lack of a similarity index for comprehensively analyzing the characteristics of seals in multiple dimensions in the prior art, the present invention provides a method and system for identifying the authenticity of seals on calligraphy and paintings.

[0006] To achieve the above technical objectives, the technical solutions adopted by the present invention are as follows:

[0007] A method for identifying the authenticity of seals on calligraphy and paintings, comprising the steps of:

[0008] S1. Collect the image information of the original work and establish a corresponding database of the seal and the original work;

[0009] S2. Construct a seal portrait from three dimensions: the overall characteristics, local characteristics, and associated characteristics of the seal;

[0010] S3. Construct the association relationship between the seals in the work, and combine the collection seal with the existing work age information to obtain the total seal map, relationship map, clustering map, and age time axis;

[0011] S4. The comprehensive similarity index of the seal is formed according to the weighted average of the three dimensions, and the authenticity of the seal work is comprehensively judged.

[0012] Further, the overall characteristics include the low-dimensional characteristics and high-dimensional characteristics of the seal, representing the overall information of the seal;

[0013] The local characteristics include regional characteristics, corner point characteristics, and contour characteristics. The regional characteristics are used to represent the trend of the seal, the corner point characteristics are used to represent the information of the seal stroke edges and corners, and the contour characteristics are the outer contour information of the seal, used to represent the geometric shape of the seal;

[0014] The associated characteristics combine the similarity information of all other seals of the same author in the database where the seal is located to assist in representing the similarity of the seal.

[0015] Further, the extraction method of the overall characteristics includes:

[0016] S101. Use the kmeans clustering method to extract the foreground of the seal and resize the seal foreground to a size of 224*224 for subsequent processing;

[0017] S102. We crop multiple images of size 224*224 from the calligraphy and painting works as the background;

[0018] S103. Combine the obtained background image with the seal foreground to form new seal data;

[0019] S104. We train a seal feature extraction model to extract the two-dimensional features of all seals of the same author and perform standardization and normalization processing.

[0020] Furthermore, when comparing the overall features, the cosine of the two-dimensional feature vectors of the two seals is used as the similarity of the overall features; let the two seal images S i and S j , and their overall feature vectors f i and f j are extracted respectively. Then their overall feature similarity F G can be expressed as:

[0021]

[0022] Furthermore, the Harris algorithm is used to extract the corner features. For the cropped seal, the foreground of the seal obtained by kmeans clustering is used to extract the corner features. When comparing, the ratio of the number of corner features to the number of corner features of the standard seal is used as the similarity of the corner features;

[0023] Let the two seal images S i and S j , and their corner features N ci and N cj are extracted respectively. Then their corner feature similarity F C is expressed as:

[0024]

[0025] The contour feature extraction is carried out on the basis of the foreground of the cropped seal obtained by kmeans clustering;

[0026] First, traverse the foreground pixels of the seal from top to bottom to find the possible contour points of the upper and lower boundaries, then traverse from left to right and record the contour points of the left and right boundaries. Finally, use the convex hull method to draw the overall contour; when comparing, calculate the discrete Fréchet distance of the contour curves of the two seals as the similarity of the contour features;

[0027] Let the contour feature f i of the seal image S i satisfy f i : [a, b] → V, and the contour feature f j of the seal image S j satisfy f j : [a, b] → V, where V is the set of album collections of the same author in the database; then their contour feature similarity F P is expressed as:

[0028]

[0029] Furthermore, the SIFT algorithm is used to extract the region features. The foreground and background of the seal are regarded as a whole, and the SIFT features are directly extracted from the cropped seal;

[0030] During the comparison, the SIFT features of the two seals are matched, and the ratio of the number of matching feature points obtained to the total number of feature points of the standard seal is used as the similarity of the local features. Let the two seal images be S i and S i , the number of regional features be N ri and N rj , and the number of matching feature points be M rij , then the regional feature similarity F R is expressed as:

[0031]

[0032] Furthermore, the associated feature is the weighted feature mean of the overall feature F G , corner feature F C , contour feature F P and regional feature F R of other seals in the same album set by the same author in the database;

[0033] For each seal image S i , the four-dimensional feature mean M ij with the original seal can be expressed as:

[0034] M ij = F Gij ·β1 + F Cij ·β2 + F Pij ·β3 + F Rij ·β4

[0035] where j ∈ [0, T], j ≠ i, and β1, β2, β3, β4 are weight coefficients, all set to 0.25; the associated feature F i of seal S Ai can be expressed as:

[0036] F Ai = ∑ j∈(0，K]，j≠i M ij .

[0037] Furthermore, the overall feature F i , corner feature F Gi , contour feature F Ci , regional feature F Pi , and associated feature F Ri obtained for each seal image S Ai , and the similarity Sim j with the original seal S ij is expressed as:

[0038] Sim ij = FGij ·ω1 + F Cij ·ω2 + F Pij ·ω3 + F Rij ·ω4 + F Aij ·ω5

[0039] where j ≠ i, ω1, ω2, ω3, ω4, ω5 are weight coefficients, all set to 0.2.

[0040] Furthermore, use the labelme tool to label the seals therein and rotate them by a certain angle θ, where θ ∈ [-45°, 45°].

[0041] A system for identifying the authenticity of seals on calligraphy and paintings includes an acquisition unit, a data storage unit, a unit for constructing a seal feature portrait, a unit for comprehensively analyzing the work age information, and a comprehensive judgment unit;

[0042] The acquisition unit acquires the original work image information;

[0043] The data storage unit is used to store all the original works of the seals of the same author in the same album set according to the corresponding relationship;

[0044] The unit for constructing a seal feature portrait constructs a seal portrait from three dimensions: the overall features, local features, and associated features of the seal;

[0045] The unit for comprehensively analyzing the work age information analyzes the features of the same type of seals, the differences and connections between this seal and the same type of seals through the seal general atlas, relationship atlas, clustering atlas, and age time axis;

[0046] The comprehensive judgment unit constitutes the comprehensive similarity index of the seal according to the weighted average of the three dimensions, and comprehensively judges the authenticity of the seal work.

[0047] Compared with the prior art, the present invention has the following beneficial effects:

[0048] It breaks the form that the seal and its source work must be viewed in a bound manner, extracts the seals in the current scattered calligraphy and painting works, corresponds them to their source works, and establishes a database for easy query and retrieval;

[0049] It establishes the association relationship between seals, which is convenient for comprehensive analysis of seals;

[0050] It integrates other information such as collection seals and ages in the works, and establishes multiple views to assist in the analysis of seals;

[0051] It improves the current seal comparison method and improves the accuracy of the analysis tool. BRIEF DESCRIPTION OF THE DRAWINGS

[0052] Figure 1It is a flowchart of a method for distinguishing the authenticity of seals on calligraphy and paintings of the present invention;

[0053] Figure 2 It is a specific step flowchart of a method for extracting the overall features of the present invention. Specific implementation manner

[0054] For the convenience of understanding by those skilled in the art, the present invention will be further described below in conjunction with embodiments and the accompanying drawings. The content mentioned in the implementation manner does not limit the present invention.

[0055] Such as Figure 1 As shown, this embodiment provides a method for distinguishing the authenticity of seals on calligraphy and paintings, including the steps: S1, collecting the image information of the original work and establishing a corresponding database of the seal and the original work; S2, constructing a seal portrait from three dimensions: the overall features, local features, and associated features of the seal; S3, constructing the associated relationship between the seals in the work, and combining the collection seal with the existing work age information to obtain the seal total map, relationship map, clustering map, and age timeline.

[0056] To comprehensively compare and analyze the seal by integrating other information of the seal, we provide a semi-automated auxiliary analysis tool. This tool consists of four parts: the seal total map, relationship map, clustering map, and age timeline. It constructs the associated relationship between the seals, integrates the collection seal and work age information and presents them well, so as to assist experts in more comprehensively comparing and analyzing the seals. Among them, the seal total map is the overall relationship map of all the seals in Shi Tao's paintings, which shows the distribution of the seals in Shi Tao's different paintings; the relationship map is an overview of the appearance of the seals with the same interpretation in all of Shi Tao's paintings, which shows the frequency of the appearance of such seals in Shi Tao's different works and the connection between them, facilitating the analysis of the frequency of use of this seal in a certain period and the evolution of the same type of seals in different paintings; the clustering map is an overview of the clustering arrangement of the seals with the same interpretation according to similarity, which is a prominent presentation of different forms of such seals, facilitating the grasping of the characteristics of the same type of seals, and then analyzing the differences and connections between this seal and the same type of seals; the age timeline is the arrangement of the appearance of Shi Tao's seals in works of different ages, facilitating the analysis of whether the appearance time of this seal conforms to historical records by combining the work age information and the earliest appearance time information of the seal.

[0057] S4, constituting the comprehensive similarity index of the seal according to the weighted average of the three dimensions, and comprehensively judging the authenticity of the seal work.

[0058] In step S1, to obtain the seal, the Yolov4 object detection framework is adopted to train the seal dataset, obtaining a seal detection model. The seals in multiple calligraphy and painting works by the same author are detected, and for the seals with a confidence level greater than 90%, the detection bounding boxes are expanded by 10 pixel points and cropped into squares based on the long side. Thus, the seals maintain their original forms in subsequent processing steps.

[0059] The seal information obtained in step S1 needs further processing. The original seal images obtained by cropping are mapped to the HSV space and the black parts in the seals are removed.

[0060] The global features include the low-dimensional features and high-dimensional features of the seal, representing the overall information of the seal; the local features include regional features, corner features, and contour features. The regional features are used to represent the trend of the seal, the corner features are used to represent the information of the edges and corners of the seal strokes, which are important detailed features for identification, and the contour features are the outer contour information of the seal, used to represent the geometric shape of the seal; the association features combine the similarity information of all other seals by the same author in the database where the seal is located to assist in representing the similarity of the seal.

[0061] As Figure 2 shown, the methods for extracting global features include: S101. Use the kmeans clustering method to extract the foreground of the seal and resize the seal foreground to a size of 224*224 for subsequent processing; S102. We crop multiple images of size 224*224 from the calligraphy and painting works as the background; S103. Combine the obtained background images with the seal foreground to form new seal data; S104. We train a seal feature extraction model to extract the two-dimensional features of all seals by the same author and perform standardization and normalization processing.

[0062] When comparing global features, the cosine of the two-dimensional feature vectors of the two seals is used as the similarity of the global features. Let the two seal images be S i and S j , and their global feature vectors f i and f j are extracted respectively. Then their global feature similarity F G can be expressed as:

[0063]

[0064] The extraction of corner features uses the Harris algorithm. The foreground of the cropped seal obtained by kmeans clustering of the cropped seal is used to extract corner features. When comparing, the ratio of the number of corner features to the number of corner features of the standard seal is used as the similarity of corner features. At this time, complex background corners may be recognized as seal corners, which will affect the accuracy of seal corner feature extraction. Therefore, to avoid interference from background factors, we use the foreground of the cropped seal obtained by kmeans clustering of the cropped seal to extract corner features.

[0065] Let two seal images be S i and S j . Respectively extract their corner features N ci and N cj . Then their corner feature similarity F C is expressed as:

[0066]

[0067] The extraction of contour features is carried out on the basis of the foreground of the cropped seal obtained by kmeans clustering of the cropped seal; when extracting contour features, considering that the complex background may interfere with the contour of the seal.

[0068] First, traverse the pixels of the foreground of the seal from top to bottom to find the possible contour points of the upper and lower boundaries, then traverse from left to right and record the contour points of the left and right boundaries. Finally, use the convex hull method to draw the overall contour; when comparing, calculate the discrete Fréchet distance of the contour curves of the two seals as the similarity of the contour features.

[0069] Let the contour feature f i of the seal image S i satisfy f i : [a, b] → V, and the contour feature f j of the seal image S j satisfy f j : [a, b] → V, where V is the set of album collections of the same author in the database; then their contour feature similarity F P is expressed as:

[0070]

[0071] The extraction of region features uses the SIFT algorithm. The foreground and background of the seal are regarded as a whole, and the SIFT features are directly extracted from the cropped seal; since the paper materials and preservation conditions used in the same period or the same painting are usually similar, the background factors can also be regarded as part of the seal features. We directly extract the SIFT features from the cropped seal.

[0072] During comparison, the SIFT features of the two seals are matched, and the ratio of the number of matching feature points obtained to the total number of feature points of the standard seal is used as the similarity of the local features. Let the two seal images be S i and S j , their respective numbers of regional features be N ri and N rj , and the number of their matching feature points be M rij , then the regional feature similarity F R is expressed as:

[0073]

[0074] The associated features are the weighted feature means of the overall feature F G , corner feature F C , contour feature F P and regional feature F R of other seals in the same album set by the same author in the database.

[0075] For each seal image S i , the four-dimensional feature mean M ij with the original seal can be expressed as:

[0076] M ij = F Gij ·β1 + F Cij ·β2 + F Pij ·β3 + F Rij ·β4

[0077] where j ∈ [0, T], j ≠ i, and β1, β2, β3, β4 are weight coefficients, all set to 0.25; the associated feature F i of seal S Ai can be expressed as:

[0078]

[0079] Having obtained the overall feature F i , corner feature F Gi , contour feature F Ci , regional feature F Pi , and associated feature F Ri of each seal image S Ai , the similarity Sim j with the original seal S ij is expressed as:

[0080] Sim ij = F Gij ·ω1 + F Cij ·ω2 + F Pij ·ω3 + F Rij·ω4 + F Aij ·ω5

[0081] where j ≠ i, and ω1, ω2, ω3, ω4, ω5 are weight coefficients, all set to 0.2.

[0082] Use the labelme tool to annotate the seals therein and rotate them by a certain angle θ, where θ ∈ [-45°, 45°]. Expand the dataset to enhance the diversity of the data to adapt to the situation where the seals in the works have rotation angles. In the seal editing box, perform pairwise comparison operations on the seals to be compared and their similar seals. Through functions such as dragging, scaling, and rotating, overlap the two seals visually, view the differences in the detailed parts of the seals, calculate the overlapping degree of the two seals, and further conduct comparison and analysis of the seals.

[0083] A system for discriminating the authenticity of seals on calligraphy and paintings, including a collection unit for collecting the original work image information; a data storage unit for storing all the original works of the seals of the same author in the same album set according to the corresponding relationship; a unit for constructing a seal feature portrait for constructing a seal portrait from three dimensions: the overall features, local features, and associated features of the seal; a unit for comprehensive analysis of the work age information for analyzing the features of the same type of seals, the differences and connections between this seal and the same type of seals through the seal general atlas, relationship atlas, clustering atlas, and age time axis; and a comprehensive judgment unit for forming a comprehensive similarity index of the seal according to the weighted average of the three dimensions and comprehensively judging the authenticity of the seal work.

[0084] Compared with the prior art, the present invention has the following beneficial effects:

[0085] It breaks through the form that the seal and its source work must be bound for viewing, extracts the seals in the current scattered calligraphy and painting works and corresponds them to their source works, and establishes a database for easy query and retrieval; establishes the association relationship between seals for easy comprehensive analysis of seals; integrates other information such as collection seals and ages in the works and establishes multiple views to assist in the analysis of seals; improves the current seal comparison method and improves the accuracy of the analysis tool.

[0086] The above provides a detailed introduction to a method and system for discriminating the authenticity of seals on calligraphy and paintings provided by the present application. The description of the specific embodiments is only used to help understand the method and its core idea of the present application. It should be noted that for those of ordinary skill in the art in this technical field, without departing from the principle of the present application, several improvements and modifications can be made to the present application, and these improvements and modifications also fall within the protection scope of the claims of the present application.

Claims

1. A method for distinguishing the authenticity of seals on calligraphy and paintings, characterized in that, Including the steps: S1. Collect the image information of the original work and establish a corresponding database of the seal and the original work; S2. Construct a seal portrait from three dimensions: the overall features, local features, and associated features of the seal; S3. Construct the association relationships between the seals in the work, and combine the collection seals with the existing work age information to obtain the overall seal atlas, relationship atlas, clustering atlas, and age timeline; S4. The comprehensive similarity index of the seal is formed based on the weighted average of the three dimensions, and the authenticity of the seal work is comprehensively judged; The overall features include the low-dimensional features and high-dimensional features of the seal, representing the overall information of the seal; The local features include regional features, corner features, and contour features. The regional features are used to represent the trend of the seal, the corner features are used to represent the information of the seal stroke edges and corners, and the contour features are the outer contour information of the seal, used to represent the geometric shape of the seal; The associated features combine the similarity information of all other seals of the same author in the database where the seal is located to assist in representing the similarity of the seal; The harris algorithm is used to extract the corner features. For the cropped seal, the corner features are extracted from the foreground of the seal obtained by kmeans clustering. When comparing, the ratio of the number of corner features to the number of corner features of the standard seal is used as the similarity of the corner features; Let two seal images be S i and S j . Respectively extract their corner features N ci and N cj . Then their corner feature similarity F C is expressed as: The contour feature extraction is carried out on the basis of the foreground of the seal obtained by kmeans clustering of the cropped seal; First, traverse the foreground pixels of the seal from top to bottom to find the possible contour points of the upper and lower boundaries, then traverse from left to right to record the contour points of the left and right boundaries, and finally use the convex hull method to draw the overall contour; when comparing, the discrete Fréchet distance of the contour curves of the two seals is calculated as the similarity of the contour features; Set the seal image S i The contour feature f i Satisfies f i :[a, b] → V, the seal image S j The contour feature f j Satisfies f j :[a, b] → V, where V is the set of album collections by the same author in the database; then their contour feature similarity F P Is expressed as: When comparing, the sift features of the two seals are matched, and the ratio of the number of matching feature points obtained to the total number of feature points of the standard seal is used as the similarity of the local features; Let there be two seal images S i and S j , the number of regional features be N ri and N rj , and the number of matching feature points be M rij , then the regional feature similarity F R is expressed as: The associated feature is the overall feature F of other seals in the same album collection of the same author in the database G , the corner feature F C , the contour feature F P and the weighted feature mean of the region feature F R ; For each seal image S i , with the four-dimensional feature mean value M of the original seal ij is expressed as: M ij = F Gij ·β1 + F Cij ·β2 + F Pij ·β3 + F Rij ·β4 where \(j\in[0,T]\), \(j\neq i\), and \(\beta_1\), \(\beta_2\), \(\beta_3\), \(\beta_4\) are weight coefficients, all set to 0.25; the associated feature \(F\) of the seal \(S\) i is expressed as: Ai is denoted as: F Ai = ∑ j∈(0,K],j≠i M ij ; Each obtained seal image S i with overall feature F Gi , corner feature F Ci , contour feature F Pi , region feature F Ri and associated feature F Ai , the similarity Sim j with the original seal S ij is expressed as: Sim ij = F Gij · ω1 + F Cij · ω2 + F Pij · ω3 + F Rij · ω4 + F Aij · ω5 Where j≠i, ω1, ω2, ω3, ω4, ω5 are weight coefficients, all set to 0.

2.

2. The authenticity discrimination method of the seal on calligraphy and painting according to claim 1, characterized in that, The extraction method of the overall features includes: S101. Use the kmeans clustering method to extract the foreground of the seal and resize the seal foreground to a size of 224*224; S102. Crop multiple images of size 224*224 from the calligraphy and painting works as the background; S103. Combine the obtained background images with the seal foreground to form new seal data; S104. Train to obtain a seal feature extraction model, extract the two-dimensional features of all seals of the same author and perform standardization and normalization processing.

3. A method for distinguishing the authenticity of a seal on a painting or calligraphy according to claim 2, characterized in that, When comparing the overall features, the cosine of the two-dimensional feature vectors of the two seals is used as the similarity of the overall features; Let two seal images be S i and S j , and their overall feature vectors f i and f j are extracted respectively. Then their overall feature similarity F G is expressed as:

4. A method for identifying the authenticity of seals on calligraphy and paintings according to claim 3, characterized in that, Use the labelme tool to label the seals therein and rotate them by a certain angle θ, where θ∈[-45°, 45°].

5. A system for discriminating the authenticity of seals on calligraphy and paintings implementing the method for discriminating the authenticity of seals on calligraphy and paintings according to claim 1, characterized in that, Including a collection unit, a data storage unit, a unit for constructing a seal feature portrait, a unit for comprehensive analysis of work age information, and a comprehensive judgment unit; The collection unit collects the image information of the original work; The data storage unit is used to store all the original works of the seals of the same author in the same album set according to the corresponding relationship; Build a seal feature portrait unit to construct a seal portrait from three dimensions: the overall features, local features, and associated features of the seal; A comprehensive analysis unit for the work age information analyzes the characteristics of the same type of seals, the differences and connections between this seal and the same type of seals through the seal general atlas, relationship atlas, clustering atlas, and age timeline; A comprehensive judgment unit forms a comprehensive similarity index of this seal according to the weighted average of the three dimensions, and comprehensively judges the authenticity of the seal works.

Images