Database establishment method based on age discrimination of artworks

Through multimodal imaging technology and deep learning models, multi-dimensional features of artworks are extracted and dynamically updated databases are constructed, which solves the problem of insufficient discrimination accuracy and applicability in artwork identification, and achieves high efficiency for various artworks , the ability to accurately distinguish and adapt to new types.

CN119066216AInactive Publication Date: 2024-12-03JIANGXI CHIFANG ART APPRAISAL CO LTD
View PDF 0 Cites 4 Cited by

Patent Information

Application Number
CN202411331186.9
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2024-09-24
Publication Date
2024-12-03
Estimated Expiration
Not applicable · inactive patent

AI Technical Summary

Technical Problem

The existing art appraisal technology has problems such as limited discrimination accuracy, insufficient non-destructive and broad applicability, as well as lack of dynamic updates and adaptability. It is especially prone to misjudgment when facing counterfeits with exquisite craftsmanship and superb counterfeiting technology. It is difficult to meet the needs of large-scale art appraisal.

Method used

Using multimodal imaging technology and deep learning models, multi-dimensional data is obtained through comprehensive scanning of artworks, geometric, texture, material composition and time attenuation features are extracted, standard databases are built and real-time updates are made, and deep learning models are used to automate features. Extract and judge to achieve accurate judgment of the age of artworks.

Benefits of technology

Improve the accuracy and applicability of artwork age discrimination, and can obtain detailed information without destroying artworks. It is suitable for artwork discrimination of various types and ages, realizes efficient processing of large-scale artwork inspections, and passes The dynamic update mechanism continues to adapt to new types of artworks, enhancing the long-term application value of the system.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN119066216A_ABST
    Figure CN119066216A_ABST
Patent Text Reader

Abstract

The invention relates to the technical field of artwork identification, and discloses a database establishment method based on age discrimination of artworks, and the method comprises the following steps: S1, selecting a plurality of artwork samples with known age, type and style; s2, performing comprehensive scanning on the sample by using a multi-modal imaging technology; s3, carrying out preprocessing on the multi-modal imaging data; s4, extracting features from the preprocessed multi-modal data; s5, constructing a deep learning model; s6, storing the extracted multi-modal feature data in a standard database; s7, performing multi-modal data acquisition and feature extraction on the artwork to be identified by using the deep learning model, and comparing with data in the standard database; and S8, generating a credibility score based on a comparison result. By introducing the multi-modal imaging technology and the deep learning model, the multi-dimensional features of the artwork can be comprehensively extracted, and subtle differences which are difficult to identify by a single imaging technology are captured.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention relates to the technical field of artwork identification, and in particular to a method for establishing a database based on identifying the age of artworks. Background Art

[0002] Art authentication is a complex and highly specialized field that involves determining the age, authenticity, and historical value of artworks. Traditional methods of art authentication rely mainly on expert experience, visual observation, and literature research. These methods rely heavily on the subjective judgment of the appraiser, and have a high degree of dependence on personal experience and uncertainty, especially when faced with counterfeits with exquisite craftsmanship and advanced imitation technology, which are prone to misjudgment. In addition, traditional methods are inefficient and cannot meet the needs of large-scale art authentication.

[0003] In recent years, the progress of science and technology has gradually applied chemical analysis, spectroscopy and other methods to the identification of artworks. These technologies can accurately analyze the material composition of artworks, thereby assisting in determining their age and authenticity. However, such technologies are usually destructive and cannot be widely applied to all types of artworks. In addition, these technologies can often only provide information in a single dimension, making it difficult to fully reflect the complex characteristics of artworks, resulting in limitations in the identification results.

[0004] The shortcomings of the existing technology are mainly reflected in the following aspects: Limited accuracy of identification: Traditional visual observation and empirical identification methods are limited by the subjective judgment of experts and cannot guarantee the objectivity and consistency of identification. Existing chemical analysis and single imaging technology, although they can provide a certain degree of accuracy, are difficult to accurately distinguish artworks of different ages and craftsmanship in all cases due to the limitations of information dimensions.

[0005] Insufficient non-destructiveness and wide applicability: Although chemical composition analysis technology has certain advantages in material testing, its destructiveness to artworks limits its scope of application. In particular, the demand for non-destructive technology is more urgent when identifying precious antiques and artworks. Existing technologies can often only be used on a small number of samples and cannot achieve large-scale non-destructive testing.

[0006] Lack of dynamic updating and adaptability: Existing identification technologies and databases are usually static and difficult to update in a timely manner as new samples are added. This means that when faced with new types of artworks or crafts, existing technologies may not be able to adapt effectively, resulting in inaccurate identification results. In addition, existing technologies lack intelligent learning and optimization mechanisms, making it difficult to continuously improve their identification capabilities.

[0007] To this end, those skilled in the art provide a method for establishing a database based on the age determination of artworks to solve the problems raised in the above-mentioned background technology. Summary of the invention

[0008] In view of the shortcomings of the prior art, the present invention provides a method for establishing a database based on the age identification of artworks, which solves the problems of limited identification accuracy, lack of non-destructiveness and wide applicability, and lack of dynamic updating and adaptability in the prior art.

[0009] To achieve the above objectives, the present invention is implemented by the following technical solutions: A method for establishing a database based on the age discrimination of artworks comprises the following steps: S1. Select multiple artwork samples of known age, type, and style; S2. Performing a comprehensive scan of the sample using a multimodal imaging technology, wherein the multimodal imaging technology includes but is not limited to optical microscopy, electron microscopy, hyperspectral imaging, X-ray imaging, and infrared imaging, for obtaining multi-dimensional data of the sample; S3, preprocessing the multimodal imaging data, including image normalization, noise removal, multimodal data alignment and fusion; S4, extracting features from the preprocessed multimodal data, including one or more of geometric features, texture features, material composition features, and time decay features; S5. Build a deep learning model, including a convolutional neural network and a multimodal data fusion network, to train the extracted features; S6, storing the extracted multimodal feature data in a standard database, which is classified and organized according to the age, type, and style of the samples; S7, using the deep learning model to perform multimodal data collection and feature extraction on the artwork to be authenticated, and comparing the data with the data in the standard database; S8. Generate a credibility score based on the comparison results to determine whether the age of the artwork to be authenticated matches a certain category of samples in the standard database. As new samples are added, the data content and model training data of the standard database are updated in real time, and the deep learning model is updated through incremental learning technology.

[0010] Preferably, the multimodal data fusion in step S3 includes the following calculation steps: Image data acquired by different imaging techniques are registered and aligned, and the translation vector and rotation angle of each modality image are calculated using an image registration algorithm so that all modality images overlap in geometric space; The pixel-level fusion of the aligned images uses the weighted average method, where the fused image The calculation formula is:

[0011] in, , respectively represent the aligned images of different modalities, is the weight coefficient of each modality image, satisfying .

[0012] Preferably, the geometric feature extraction in step S4 specifically includes: The edge information in the image is extracted by morphological operation, and the edge pixel set E is calculated by the Canny edge detection algorithm; Calculate the morphological features of the edge pixel set, including the perimeter of the contour and area , and further calculate the compactness of the shape : Each extracted geometric feature is normalized and the feature value is mapped to the [0, 1] interval to ensure the comparability between different features.

[0013] Preferably, the texture feature extraction in step S4 includes the following calculation steps: The texture features of the image, including contrast, are calculated using the gray-level co-occurrence matrix ,energy Homogeneity and entropy , and the calculation formulas are:

[0014]

[0015]

[0016]

[0017] in, Represents the elements in the co-occurrence matrix The probability value of .

[0018] Preferably, the material composition feature extraction in step S4 includes: Based on the hyperspectral imaging data, the spectral data is reduced in dimension through principal component analysis to extract the main component features; The spectral curve of the hyperspectral image under the main components is calculated, and the characteristic vectors of different materials are calculated based on the reflectivity of specific bands.

[0019] Preferably, the time decay feature extraction in step S4 comprises the following steps: Analyze the oxidation degree of sample materials and calculate the oxide concentration using X-ray imaging data; Calculate the color change of the material surface and evaluate the aging rate of the material based on time series analysis of hyperspectral data.

[0020] Preferably, the specific steps of deep learning model training in step S5 are: The extracted geometry, texture, material composition, and time decay features are passed as input feature vector x to the convolutional neural network; The model parameters are updated through the back-propagation algorithm, the prediction error is calculated using the cross entropy loss function, and the loss function is minimized to optimize the model:

[0021] in, is the true label, is the model prediction value, is the model parameter.

[0022] Preferably, in step S6, the standard database is organized as follows: The characteristics of multimodal data are stored in layers according to the age, type, and style of the samples; Each feature set is indexed to ensure that the database can quickly compare all relevant feature data for a specific sample when searching.

[0023] Preferably, in step S8, the calculation of the credibility score includes: Compare the multimodal features of the artwork to be authenticated with the corresponding features in the standard database and calculate the Euclidean distance As a similarity measure:

[0024] Based on the similarity measurement results, the soft maximum function is used to calculate the probability of each decade category and generate a credibility score:

[0025] in, is the Euclidean distance between the sample to be identified and the feature of the jth category, is the total number of categories.

[0026] The present invention provides a method for establishing a database based on the age discrimination of artworks. It has the following beneficial effects: 1. By introducing multimodal imaging technology and deep learning models, the present invention can comprehensively extract the multi-dimensional features of artworks and capture subtle differences that are difficult to identify with a single imaging technology.

[0027] 2. The multimodal imaging method of the present invention can obtain detailed internal and surface information without destroying the artwork, and is suitable for the identification of artworks of various types and ages. In addition, the deep learning model can efficiently process a large number of samples through automated feature extraction and identification, and is suitable for large-scale artwork detection and identification.

[0028] 3. The present invention introduces a dynamic update mechanism and incremental learning technology of the standard database, which can update the data content and model parameters in real time as new samples are added. This mechanism ensures that the system can continuously adapt and optimize, and still maintain efficient and accurate discrimination capabilities for new types of artworks, thereby enhancing the long-term application value of the system. BRIEF DESCRIPTION OF THE DRAWINGS

[0029] Figure 1 It is a schematic diagram of the method flow of the present invention. DETAILED DESCRIPTION

[0030] The technical solution of the present invention will be described clearly and completely below in conjunction with the accompanying drawings 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.

[0031] Example 1: Multimodal imaging and database establishment Sample selection Five specimens of blue and white porcelain from the Ming Dynasty were selected as samples. These specimens were confirmed by cultural relic identification experts and third-party testing agencies to be authentic products from the Ming Dynasty. These samples were selected from different types and styles to ensure diversity.

[0032] On each piece of porcelain selected, 10 standard points were randomly selected. These standard points are distributed in different parts of the porcelain, such as the bottle mouth, bottle body, base, etc.

[0033] Application of multimodal imaging technology Optical microscope: Use an optical microscope to perform preliminary surface microstructure imaging of each standard point to obtain the surface texture and crack conditions of the sample.

[0034] Electron microscope: Based on the optical microscope, the electron microscope is further used to perform higher resolution imaging at standard points to capture the material structure and process traces at the microscopic level.

[0035] Hyperspectral imaging: Hyperspectral imaging is performed at each standard point to obtain the spectral information at that point. By analyzing the spectral data, the pigment composition, firing temperature and changes in the surface enamel of the porcelain can be determined.

[0036] X-ray imaging: Use X-rays to scan the internal structure of porcelain and analyze information such as the density distribution of its body and internal cracks.

[0037] Data preprocessing Image normalization: Normalize all images and adjust pixel values ​​to the same range (0-1) to eliminate differences caused by different imaging devices.

[0038] Noise removal: Use denoising algorithms such as Gaussian filtering to remove random noise in the image and retain key structural information.

[0039] Multimodal data alignment and fusion: SIFT feature point matching algorithm is used to align images under different modalities to ensure the consistency of all images in spatial position. The fused images are processed using the weighted average method. The weight coefficients of each modality image are determined through experiments, which are: optical microscope 0.2, electron microscope 0.3, hyperspectral imaging 0.3, and X-ray imaging 0.2.

[0040] Feature extraction Geometric feature extraction: Use the Canny edge detection algorithm to extract the edge features of the porcelain surface and calculate the perimeter and area of ​​its contour.

[0041] Texture feature extraction: The texture features of the porcelain surface, such as contrast, energy, homogeneity, and entropy, were extracted based on gray-level co-occurrence matrix (GLCM) analysis.

[0042] Material composition feature extraction: By performing principal component analysis (PCA) on hyperspectral imaging data, the main spectral feature vectors are extracted to reflect the pigment and glaze composition of the porcelain surface.

[0043] Time decay feature extraction: By analyzing X-ray imaging data, the degree of oxidation inside the porcelain is calculated. Combined with hyperspectral imaging data, color changes are analyzed to evaluate the aging speed and age characteristics of the porcelain.

[0044] Standard database establishment The multimodal feature data extracted from the 10 standard points of the five pieces of porcelain are stored in the database. The feature data of each piece of porcelain is classified by type, age, style, and annotated with corresponding metadata.

[0045] The standard database will eventually include 20,000 feature vectors (5 pieces of porcelain, 10 standard points, and 400 features extracted from each point), which will serve as comparison standards for future artworks to be authenticated.

[0046] Example 2: Artwork Age Identification Based on Multimodal Machine Learning Multimodal imaging of samples to be identified A piece of porcelain of unknown age was subjected to the same multimodal imaging process as in Example 1. Five standard points of the porcelain were selected and comprehensively scanned by optical microscopy, electron microscopy, hyperspectral imaging, and X-ray imaging.

[0047] Feature extraction and database comparison The corresponding geometry, texture, material composition and time decay features are extracted from the multimodal data of the sample to be identified, and a feature vector is generated.

[0048] Using the trained deep learning model, the extracted feature vector is compared with the feature data in the standard database to calculate the similarity of each standard point. The similarity is calculated using the Euclidean distance:

[0049] in, is the feature vector of the sample to be identified, is the feature vector of the corresponding category in the database.

[0050] Credibility Scoring and Judgment Based on the comparison results, the credibility scores of the samples to be identified and the samples of each category are calculated. The probability of each category is calculated using the soft maximum function:

[0051] in, is the Euclidean distance between the sample to be identified and the feature of the jth category, Score its credibility. If the credibility score of a category If the probability is greater than 80%, it is determined that the sample to be identified and the samples of the same category in the database belong to the same era.

[0052] Example 3: Dynamic update of standard database and model New samples added and database updated When the new porcelain sample is confirmed to be authentic, the same multimodal imaging and feature extraction method as in Example 1 is used to collect data on the sample.

[0053] The multimodal feature vector Xnew of the new sample is added to the standard database. The updated database includes the data of the new sample and is reorganized and classified according to the sample age, type, and style.

[0054] Incremental learning and model updating Using incremental learning techniques, the existing deep learning model is fine-tuned based on the newly added sample data. The fine-tuning process is performed only on new data to maintain the knowledge of the original model while adapting to the new sample data.

[0055] Whenever the database is updated, the model will automatically receive new training data and update parameters. The specific steps are:

[0056] in, are the updated model parameters, is the learning rate, is the gradient of the loss function based on the new data.

[0057] Continuous optimization and model persistence As new data is continuously added, the database and model will continue to be optimized to ensure the system's efficient recognition of new samples.

[0058] The updated model will be saved as the latest version and used in subsequent artwork age determination tasks.

[0059] Example 4: Comparative experiment In order to verify the effectiveness of the method described in the present invention, the following comparative experiment is set up.

[0060] Experimental samples Thirty porcelain samples of known age were used in the experiment, including 15 blue-and-white porcelains from the Ming Dynasty and 15 porcelains from the Qing Dynasty. The authenticity and age of each sample have been confirmed by an authoritative organization.

[0061] Ten standard points were selected for each sample, and data were collected using the multimodal imaging technology of the present invention and the traditional single imaging technology.

[0062] Experimental setup Comparison Group 1: Traditional Single Imaging Technology The samples were imaged using only an electron microscope to extract geometric and textural features of a single modality.

[0063] The traditional feature matching algorithm is used for discrimination, and the constructed database only includes the feature data of a single modality.

[0064] Comparative group 2: the method of the present invention The multimodal imaging technology mentioned in the present invention, including optical microscopy, electron microscopy, hyperspectral imaging, X-ray imaging, etc., is used to extract multimodal features.

[0065] A deep learning model is used to train and compare multimodal features, and the established standard database contains multimodal feature data.

[0066] Experimental procedures For each porcelain sample, data collection and analysis were performed using the methods of comparison group 1 and comparison group 2 respectively.

[0067] Compare the experimental results with samples of known age and calculate the discrimination accuracy.

[0068] Experimental Results Comparison Group 1: Traditional Single Imaging Technology The accuracy of discrimination is 70%, of which the accuracy of Ming Dynasty samples is 75% and that of Qing Dynasty samples is 65%. Some samples are misjudged due to feature overlap, especially in samples with similar texture features, where the accuracy is low.

[0069] Comparative group 2: the method of the present invention The discrimination accuracy is 95%, of which the accuracy of Ming Dynasty samples is 97% and that of Qing Dynasty samples is 93%. Due to the introduction of multimodal features, this method can capture more subtle differences, especially in the material composition and time decay characteristics of different ages.

[0070] Conclusion: Experimental results show that the method for establishing a database for identifying the age of artworks combined with multimodal machine learning using the method of the present invention can significantly improve the accuracy of identification. Especially when processing samples with similar appearance features, the multimodal feature extraction and deep learning analysis of this method provide stronger differentiation capabilities.

[0071] Although embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions and variations may be made to the embodiments without departing from the principles and spirit of the present invention, and that the scope of the present invention is defined by the appended claims and their equivalents.

Claims

1. A method for establishing a database based on the age discrimination of artworks, characterized in that: The following steps are involved: S1. Select multiple artwork samples of known age, type, and style; S2. Performing a comprehensive scan of the sample using a multimodal imaging technology, wherein the multimodal imaging technology includes but is not limited to optical microscopy, electron microscopy, hyperspectral imaging, X-ray imaging, and infrared imaging, for obtaining multi-dimensional data of the sample; S3, preprocessing the multimodal imaging data, including image normalization, noise removal, multimodal data alignment and fusion; S4, extracting features from the preprocessed multimodal data, including one or more of geometric features, texture features, material composition features, and time decay features; S5. Build a deep learning model, including a convolutional neural network and a multimodal data fusion network, to train the extracted features; S6, storing the extracted multimodal feature data in a standard database, which is classified and organized according to the age, type, and style of the samples; S7, using the deep learning model to perform multimodal data collection and feature extraction on the artwork to be authenticated, and comparing the data with the data in the standard database; S8. Generate a credibility score based on the comparison results to determine whether the age of the artwork to be authenticated matches a certain category of samples in the standard database. As new samples are added, the data content and model training data of the standard database are updated in real time, and the deep learning model is updated through incremental learning technology.

2. The method for establishing a database based on identifying the age of artworks according to claim 1, characterized in that: The multimodal data fusion in step S3 includes the following calculation steps: Image data acquired by different imaging techniques are registered and aligned, and the translation vector and rotation angle of each modality image are calculated using an image registration algorithm so that all modality images overlap in geometric space; The pixel-level fusion of the aligned images uses the weighted average method, where the fused image The calculation formula is: in, , respectively represent the aligned images of different modalities, is the weight coefficient of each modality image, satisfying .

3. The method for establishing a database based on identifying the age of artworks according to claim 1, characterized in that: The geometric feature extraction in step S4 specifically includes: The edge information in the image is extracted by morphological operation, and the edge pixel set E is calculated by the Canny edge detection algorithm; Calculate the morphological features of the edge pixel set, including the perimeter of the contour and area , and further calculate the compactness of the shape : Each extracted geometric feature is normalized and the feature value is mapped to the [0, 1] interval to ensure the comparability between different features.

4. The method for establishing a database based on identifying the age of artworks according to claim 1, characterized in that: The texture feature extraction in step S4 includes the following calculation steps: The texture features of the image, including contrast, are calculated using the gray-level co-occurrence matrix ,energy , Homogeneity and entropy , and the calculation formulas are: in, Represents the elements in the co-occurrence matrix The probability value of .

5. The method for establishing a database based on identifying the age of artworks according to claim 1, characterized in that: The material composition feature extraction in step S4 includes: Based on the hyperspectral imaging data, the spectral data is reduced in dimension through principal component analysis to extract the main component features; The spectral curve of the hyperspectral image under the main components is calculated, and the characteristic vectors of different materials are calculated based on the reflectivity of specific bands.

6. The method for establishing a database based on identifying the age of artworks according to claim 1, characterized in that: The time decay feature extraction in step S4 includes the following steps: Analyze the oxidation degree of sample materials and calculate the oxide concentration using X-ray imaging data; Calculate the color change of the material surface and evaluate the aging rate of the material based on time series analysis of hyperspectral data.

7. The method for establishing a database based on identifying the age of artworks according to claim 1, characterized in that: The specific steps of deep learning model training in step S5 are: The extracted geometry, texture, material composition, and time decay features are passed as input feature vector x to the convolutional neural network; The model parameters are updated through the back-propagation algorithm, the prediction error is calculated using the cross entropy loss function, and the loss function is minimized to optimize the model: in, is the true label, is the model prediction value, is the model parameter.

8. The method for establishing a database based on identifying the age of artworks according to claim 1, characterized in that: In step S6, the standard database is organized as follows: The characteristics of multimodal data are stored in layers according to the age, type, and style of the samples; Each feature set is indexed to ensure that the database can quickly compare all relevant feature data for a specific sample when searching.

9. The method for establishing a database based on identifying the age of artworks according to claim 1, characterized in that: In step S8, the calculation of the credibility score includes: Compare the multimodal features of the artwork to be authenticated with the corresponding features in the standard database and calculate the Euclidean distance As a similarity measure: Based on the similarity measurement results, the soft maximum function is used to calculate the probability of each decade category and generate a credibility score: in, is the Euclidean distance between the sample to be identified and the feature of the jth category, is the total number of categories.

Citation Information

Cited By

  • Multi-modal analysis image identification system and method

    CN119832570A

  • Antique age determination method and system based on artificial intelligence and big data

    CN120495797A

  • Porcelain fingerprint verification method and device and storage medium

    CN120808349A

  • Methods, devices and storage media for fingerprint verification of porcelain

    CN120808349B