Method and system for extracting and classifying blue and white porcelain ornamentation

Abstract

The invention discloses a method and system for extracting and classifying blue and white porcelain patterns, and the method comprises the steps: obtaining a two-dimensional plane image of the blue and white porcelain patterns, and obtaining a source image; carrying out background separation on the source image by using a GrabCut algorithm to obtain an ornamentation image; judging whether the ornamentation image is adhered or not, and if so, removing background interference by marking the foreground and the background of the ornamentation image; s3, sending the ornamentation image output in the step S3 into a pre-trained deep learning network model to obtain a classification corresponding to the ornamentation image; and storing the ornamentation image and the classification corresponding to the ornamentation image in a visual mode. Based on an image processing algorithm and a deep learning technology, automatic extraction and efficient classification of blue and white porcelain ornaments are realized. Through a friendly user interaction interface, the room can efficiently complete loading, processing, classification and result storage of the ornamentation, and the problems that in the prior art, the ornamentation extraction precision is insufficient, the classification efficiency is low, and operation is complex are solved.

Description

Technical Field

[0001] The present invention relates to the cross-technical field of computer vision and artificial intelligence, and in particular to a method and system for extracting and classifying blue and white porcelain patterns. Background Art

[0002] Blue and white porcelain is a treasure of Chinese ceramic art, and is world-renowned for its unique underglaze painting technique and rich and diverse decorative expressions. Blue and white porcelain decorative patterns cover a variety of themes such as figures, animals, landscapes, and plants, reflecting the profound heritage and artistic essence of Chinese traditional culture. With the rapid development of digital technology, digital processing and intelligent analysis of blue and white porcelain decorative patterns have become an important direction for cultural heritage protection. However, in practical applications, the extraction and classification of blue and white porcelain decorative patterns are somewhat difficult.

[0003] Blue and white porcelain patterns are usually outlined by a single blue line, which is very close to the background color of the porcelain. In addition, the patterns may contain gradient or mottled strokes, so traditional threshold segmentation and edge detection algorithms are difficult to accurately distinguish between patterns and background. In addition, the background of different blue and white porcelains may have scratches, stains or glaze reflections, which further increases the difficulty of pattern extraction.

[0004] At the same time, there are many types of blue and white porcelain decorations, and the geometric structure, texture details and artistic style of each type of decoration are significantly different. For example: human decoration: the freehand brushwork depicts the changing human figures; animal decoration: common birds and beasts, with diverse shapes and high details; landscape decoration: natural landscapes are painted with thick and thin brushstrokes; plant decoration: including flowers, plants and trees, usually in symmetrical or continuous geometric patterns. This diversity of decorative features puts higher requirements on the extraction and classification algorithms of decorations, and it is necessary to take into account the accurate extraction and classification accuracy of different types of decorations.

[0005] Currently, the digital processing of blue and white porcelain decoration mostly uses manual annotation and traditional image processing technology. However, manual annotation is time-consuming and laborious, and there are subjective biases, which cannot meet the needs of large-scale data processing; traditional image processing methods (such as edge detection and morphological processing) have poor adaptability to complex backgrounds and it is difficult to achieve high accuracy in the processing of decoration details. Summary of the invention

[0006] The technical problem to be solved by the present invention is to provide a method and system for extracting and classifying blue and white porcelain patterns in view of the deficiencies in the above-mentioned prior art, so as to solve the problems mentioned in the above-mentioned background technology.

[0007] In order to solve the above technical problems, the technical solution adopted by the present invention is: a method and system for extracting and classifying blue and white porcelain patterns, wherein a method for extracting and classifying blue and white porcelain patterns comprises:

[0008] S1: Acquire a two-dimensional plane image of blue and white porcelain decoration to obtain a source image;

[0009] S2: Using GrabCut algorithm to separate the background of the source image to obtain a pattern image;

[0010] S3: determining whether the pattern image is adhered, and if so, removing background interference by marking the foreground and background of the pattern image;

[0011] S4: Send the pattern image outputted in step S3 to a pre-trained ResNet50 deep learning network model to obtain a classification corresponding to the pattern image;

[0012] S5: storing the pattern images and their corresponding categories in a visual manner;

[0013] The ResNet50 deep learning network model takes the pattern image as input and the classification corresponding to the pattern image as output for training.

[0014] Further, after step S1 and before step S2, the method further includes:

[0015] The source image is smoothed by using the RTV algorithm.

[0016] Furthermore, the source image is smoothed by using the RTV algorithm, including:

[0017] Performing pixel normalization processing on the source image;

[0018] Calculate the relative total variation and local total variation of the source image after normalization, and determine the smoothness of each pixel point of the source image;

[0019] The source image is smoothed according to the smoothness of each pixel and the set objective function.

[0020] Further, in step S2, the GrabCut algorithm is used to separate the background of the source image to obtain a pattern image, including:

[0021] Initializing the source image by stroking;

[0022] According to each pixel point in the source image, an energy function of the source image is established;

[0023] The source image is cut using graph cutting technology to minimize the energy function, and the cut source image is output to obtain the pattern image.

[0024] Furthermore, the energy function consists of a data term and a smoothing term.

[0025] Further, in step S1, the acquisition of the source image includes:

[0026] Acquire a blue and white porcelain image, wherein the blue and white porcelain image is a two-dimensional plane image;

[0027] A decorative area in the blue and white porcelain image is selected as the source image according to user interaction.

[0028] Based on the inventive concept of the above method, the present invention also proposes a system for extracting and classifying blue and white porcelain patterns, comprising:

[0029] An image acquisition module is used to acquire a two-dimensional plane image of the blue and white porcelain decoration to obtain a source image;

[0030] A pattern separation module, used for separating the background of the source image using a GrabCut algorithm to obtain a pattern image;

[0031] An interference removal module is used to determine whether the pattern image is adhered, and if so, to remove background interference by marking the foreground and background of the pattern image;

[0032] The pattern classification module is used to send the pattern image output by the interference removal module to the pre-trained ResNet50 deep learning network model to obtain the classification corresponding to the pattern image;

[0033] A pattern storage module is used to store pattern images and their corresponding categories in a visual manner;

[0034] The ResNet50 deep learning network model takes the pattern image as input and the classification corresponding to the pattern image as output for training.

[0035] Compared with the prior art, the present invention has the following advantages:

[0036] The present invention provides a method and system for extracting and classifying blue and white porcelain patterns, which realizes the automatic extraction and efficient classification of blue and white porcelain patterns based on image processing algorithm (GrabCut) and deep learning technology (ResNet50). Through a friendly user interaction interface, the room can efficiently complete the loading, processing, classification and result preservation of patterns, solve the problems of insufficient pattern extraction accuracy, low classification efficiency and complex operation in the prior art, and provide technical support for the digital protection and intelligent analysis of cultural heritage.

[0037] The technical solution of the present invention is further described in detail below through the accompanying drawings and embodiments. BRIEF DESCRIPTION OF THE DRAWINGS

[0038] Figure 1 A schematic diagram of the overall process of a method for extracting and classifying blue and white porcelain patterns provided by the present invention.

[0039] Figure 2 It is a structural diagram of the ResNet50 deep learning network model in the present invention.

[0040] Figure 3 It is a schematic diagram of the pattern extraction effect of the present invention.

[0041] Figure 4 It is a schematic diagram of the classification effect of the present invention.

[0042] Figure 5 Schematic diagram of the user interface of the present invention.

[0043] Figure 6 This is a schematic diagram of the system architecture design of the present invention. DETAILED DESCRIPTION

[0044] like Figure 1-6 As shown, the present invention provides a method and system for extracting and classifying blue and white porcelain patterns, wherein a method for extracting and classifying blue and white porcelain patterns comprises:

[0045] S1: Acquire a two-dimensional plane image of blue and white porcelain decoration to obtain a source image;

[0046] S2: Using GrabCut algorithm to separate the background of the source image to obtain a pattern image;

[0047] S3: determining whether the pattern image is adhered, and if so, removing background interference by marking the foreground and background of the pattern image;

[0048] S4: Send the pattern image outputted in step S3 to a pre-trained ResNet50 deep learning network model to obtain a classification corresponding to the pattern image;

[0049] S5: storing the pattern images and their corresponding categories in a visual manner;

[0050] The ResNet50 deep learning network model takes the pattern image as input and the classification corresponding to the pattern image as output for training.

[0051] Based on image processing algorithm (GrabCut) and deep learning technology (ResNet50), this paper realizes the automatic extraction and efficient classification of blue and white porcelain patterns. Through a friendly user interface, the system can efficiently complete the loading, processing, classification and result preservation of patterns, solving the problems of insufficient pattern extraction accuracy, low classification efficiency and complex operation in the prior art, and providing technical support for the digital protection and intelligent analysis of cultural heritage.

[0052] When implemented, the present invention provides a system structure as follows: Figure 6 As shown below: (1) User interaction interface module:

[0053] It provides an intuitive and easy-to-use graphical interface that supports users to load image folders, browse and select blue and white porcelain images to be processed, and select the decorative area of ​​interest through mouse interaction.

[0054] (2) Pattern extraction module:

[0055] The GrabCut algorithm is used for preliminary segmentation, and the user is allowed to optimize the extraction effect by marking the foreground and background areas to ensure accurate separation of the pattern area and the background.

[0056] (3) Pattern classification module:

[0057] Based on the ResNet50 deep learning network, the extracted pattern images are subjected to multi-classification tasks, supporting the classification of common blue and white porcelain patterns (such as figures, animals, landscapes, and plants), and outputting the classification results.

[0058] (4) Result saving module:

[0059] Supports users to save extraction and classification results as images for further analysis and presentation.

[0060] For the above steps S1-S5, specifically:

[0061] In step S1, the present invention supports the user to load a picture folder, browse and select the blue and white porcelain image to be processed, and select the decorative area of ​​interest through mouse interaction.

[0062] That is, in step S1, the acquisition of the source image includes:

[0063] Acquire a blue and white porcelain image, wherein the blue and white porcelain image is a two-dimensional plane image;

[0064] A decorative area in the blue and white porcelain image is selected as the source image according to user interaction.

[0065] And in the present invention, after step S1 and before step S2, it also includes:

[0066] The source image is smoothed by using the RTV algorithm.

[0067] The step of smoothing the source image by using the RTV algorithm includes:

[0068] Performing pixel normalization processing on the source image;

[0069] Calculate the relative total variation and local total variation of the source image after normalization, and determine the smoothness of each pixel point of the source image;

[0070] The source image is smoothed according to the smoothness of each pixel and the set objective function.

[0071] The relative total variation (RTV) algorithm is used for smoothing. RTV is an improved image smoothing algorithm that aims to smooth the texture in the image while retaining significant structural information (such as edges). Compared with traditional smoothing methods, important structural information may be blurred during the denoising and texture smoothing process. RTV effectively separates texture noise and structure by calculating the relative value of local total variation. The RTV smoothing algorithm used in the present invention is as follows:

[0072] (1) First, separate the pixels of the RGB channels of the color image and normalize them: standardize the image pixel values ​​to [0,1][0,1][0,1] to facilitate subsequent gradient calculation.

[0073] (2) Calculate the relative total variation. In the square area W(z) with pixel point z, calculate the total window variation TV in the x and y directions. x (W(z)) and TV y (W(z)):

[0074] TV x (W(z))=∑ (x,y)∈W(z) |I'(x,y)-I'(x+1,y)|

[0075] TV y (W(z))=∑ (x,y)∈W(z) |I'(x,y)-I'(x,y+1)|

[0076] Among them, I'(x,y) represents the gradient value of the image at the pixel point (x,y), which is usually used to describe the local changes of the image. I'(x,y+1) represents the gradient value of the adjacent pixel point (x+1,y) on the right side of the image in the horizontal direction (x direction).

[0077] (3) Calculate the local total variation: Calculate the smoothness of each pixel in the x and y directions.

[0078]

[0079] Among them TV L (z) is the local total variation of pixel z, which is used to measure the smoothness of pixel z and its surrounding area. x (W(z)) is the total variation in the x direction in the window W(z) around z. TV y (W(z)) is the total variation in the y direction in the window W(z) around z.

[0080] (4) Minimizing the objective function: In order to smooth the image while preserving the main structure of the image as much as possible, the present invention calculates the output image by minimizing the following objective function.

[0081]

[0082] Among them, λ and μ are regularization parameters and weight factors, respectively, which are used to balance the difference between the total variation and the smooth objective function f(x, y). λ is used to control the weight of the total variation. When it is larger, it emphasizes smooth texture. μ: controls the sensitivity of the objective function to noise and structural differences. TV L (I) is the local total variation of image I, f(x,y) is the objective function, image I is the image to be optimized, which represents the expected value of the image at the pixel position, which can be a smoothed version of the original image or an image containing noise, it represents the expected value of the image at (x,y), which can be a smoothed version of the original image or the noise level, σ is the standard deviation of the Gaussian filter, which is used to define the smoothness of f(x,y), minJ(I) is the optimized value of the objective function, which is used to generate the smoothed output image I.

[0083] In step S2, the decorative pattern is separated, that is, in step S2, the GrabCut algorithm is used to separate the background of the source image to obtain the decorative image, including:

[0084] Initializing the source image by stroking;

[0085] According to each pixel point in the source image, an energy function of the source image is established;

[0086] The source image is cut using graph cutting technology to minimize the energy function, and the cut source image is output to obtain the pattern image.

[0087] The energy function consists of a data term and a smoothing term.

[0088] 2. Pattern extraction and optimization

[0089] The present invention introduces the GrabCut algorithm in the aspect of pattern extraction. The GrabCut algorithm is an interactive foreground extraction algorithm based on image segmentation, which combines graph cuts and Gaussian mixture models (GMM) to separate the foreground and background by minimizing the energy function.

[0090] The GrabCut algorithm process adopted by the present invention is as follows:

[0091] (1) User interaction: The user specifies the foreground area in the image by drawing a rectangular box and drawing a stroke between the foreground and background.

[0092] (2) Initialization: The algorithm initializes the foreground and background estimation based on the rectangular box and stroke drawn by the user.

[0093] (3) Energy function:

[0094] The goal of GrabCut is to minimize an energy function Ε, which consists of two parts: data term and smoothness term.

[0095] The total energy function is:

[0096]

[0097] Among them, P represents the set of all pixels in the image, N represents the set of pixel p and its neighboring pixels q, and D p (F) reflects the degree to which pixel p is closer to the foreground model, such as the difference between the pixel color or brightness and the foreground statistical distribution. and D p (B) reflects the degree to which pixel p is closer to the background model, such as the difference between the pixel color or brightness and the background statistical distribution. V represents the cost of pixel p belonging to the foreground and background respectively. pq is a smoothing term, which indicates the similarity between pixel p and its neighboring pixel q. λ is the weight of the smoothing term, which controls the balance between the data term and the smoothing term. E is the total energy function, which is used to measure the quality of the segmentation result. It comprehensively considers the data term (the cost of a pixel belonging to the foreground or background) and the smoothing term (the smoothness of the boundary of the segmentation result). By minimizing E, the optimal segmentation of the foreground and background in the image can be found.

[0098] (4) The algorithm uses graph cut technology to minimize the energy function. It treats each pixel in the image as a node in the graph, and the connections between pixels represent the similarities between them. The goal of the algorithm is to minimize the "cut" in the graph, that is, the boundary separating the foreground and background.

[0099] In step 3, the steps of extracting blue and white porcelain patterns in the present invention are as follows:

[0100] (1) Initial segmentation: The user selects the area of ​​interest with the mouse, and the system initializes the GrabCut segmentation algorithm based on the rectangular selection to perform a preliminary division of the area into foreground and background.

[0101] (2) User interaction optimization: Users can further optimize the segmentation results by marking the foreground and background masks, and the system will update the pattern extraction results in real time. This process can be iterated multiple times until the pattern edges and details fully meet the user's needs.

[0102] (3) Background removal: The extracted pattern area will be saved in the form of a transparent background or a uniform black background, thereby eliminating the interference of the background on the pattern analysis.

[0103] In step 4, the patterns are classified.

[0104] In the classification process, the present invention adopts the ResNet50 deep learning model and performs the following optimizations based on the characteristics of blue and white porcelain decoration:

[0105] (1) Transfer learning: Use the pre-trained ResNet50 model, freeze some convolutional layer weights, and only adjust the fully connected layer to make it suitable for the classification task of blue and white porcelain patterns.

[0106] (2) Data enhancement: In view of the limited data set of blue and white porcelain patterns, data enhancement techniques such as random cropping, rotation, and flipping are used to expand the training set, thereby improving the generalization ability of the model.

[0107] (3) Multi-classification support: The system supports multi-classification tasks and can accurately distinguish common blue and white porcelain patterns such as figures, animals, landscapes, and plants, and provide confidence output of the classification results to facilitate subsequent analysis and display.

[0108] Step S5 is the storage process, which is stored in a visual way. The operation is simple and the function is intuitive, and it mainly includes the following parts:

[0109] (1) Image loading and browsing: After selecting a folder, all images are automatically loaded and displayed in a list in the sidebar of the interface. Users can click to switch between different images.

[0110] (2) Pattern selection and annotation: It provides a rectangular selection tool and a brush tool, allowing users to enter the corresponding mode by clicking a button, select the area of ​​interest, and annotate the foreground or background.

[0111] (3) Display of segmentation and classification results: The extraction and classification results will be displayed in real time in the central canvas of the interface, and can also be saved to a specified path.

[0112] Therefore, 1. The present invention combines the GrabCut algorithm and user dynamic interactive annotation technology in pattern extraction to overcome the limitation of the traditional algorithm's unstable segmentation effect under complex backgrounds. Users can achieve dynamic adjustment by annotating the foreground and background, effectively reduce background noise interference, and significantly improve segmentation accuracy. Experimental results show that the system's segmentation accuracy for complex background blue and white porcelain images is significantly improved compared with traditional methods, and it can retain high-frequency details while ensuring the integrity of the segmentation edges, ensuring the authenticity and integrity of the patterns.

[0113] 2. High-precision classification of diverse patterns and small sample adaptation: Using the optimized ResNet50 deep learning model, combined with transfer learning and data enhancement technology, even with limited labeled data, it can still achieve high-precision classification of various patterns such as people, animals, landscapes, plants, etc., with a classification accuracy rate of over 95%. The model has good generalization ability and can be extended to the automated analysis of more complex patterns.

[0114] 3. User-friendly and modular operation interface: Through modular user interaction design, users can complete core operations such as image loading, pattern extraction and classification without any barriers. Real-time update and visual display functions not only improve user experience, but also significantly shorten the response time of operation and adjustment.

[0115] 4. Efficient and scalable data processing capabilities: supports automatic loading and processing of batch image files, suitable for archiving and research of large-scale blue and white porcelain data. Adopts hierarchical optimization strategy and lightweight model structure to make it compatible with ordinary computing devices, while supporting GPU acceleration environment, greatly improving data processing speed.

[0116] 5. Cultural heritage protection and parameterization support: The high-quality pattern data generated by the system supports the digital archiving of blue and white porcelain patterns, providing technical support for the protection and inheritance of cultural heritage. At the same time, parameterized data can be applied to pattern feature statistics, artistic style analysis and pattern restoration, providing a reliable basis for dating and restoration.

[0117] 6. Parametric modeling of patterns that supports dating and restoration: The system realizes parametric modeling of patterns by extracting geometric features, texture parameters and style of patterns. Combined with the results of pattern classification, it can help researchers infer the age and production process of artifacts based on typical pattern features. At the same time, parametric data can be used as the basis for pattern restoration, supporting digital reconstruction of missing parts generated based on pattern style, greatly improving restoration efficiency and accuracy.

[0118] In summary, this invention shows significant advantages in pattern extraction and classification accuracy, user experience optimization, system scalability and processing efficiency, providing comprehensive technical support for the research and protection of blue and white porcelain, and at the same time has broad application prospects in the field of intelligent processing of cultural heritage.

[0119] In order to implement the above method, the pattern area of ​​interest is selected through user interaction, and the background is automatically excluded to save the extracted patterns. This method can be applied to the task of blue and white porcelain pattern classification.

[0120] Specific implementation steps:

[0121] (1) Image loading and display: The user selects the blue and white porcelain image to be processed through the graphical user interface (GUI), and the interface loads and displays the image.

[0122] (2) User-selected foreground area: The user uses a rectangular frame to select the pattern area of ​​interest in the image. The selected area is identified as the foreground of the image.

[0123] (3) Background selection: Users can select the background area in the image by clicking multiple areas. After each selection, the background area will be gradually removed, leaving the foreground pattern.

[0124] (4) GrabCut algorithm processing: The system uses the GrabCut algorithm to segment the image, retain the selected foreground area, and remove the background.

[0125] (5) Pattern saving: The extracted foreground pattern will be saved as an independent image. The user can specify the saving path and the file name will be automatically generated.

[0126] (6) Interactive operation: Users can switch to different images through buttons in the interface and repeat the selection operation until all patterns are extracted.

[0127] The technical effect achieved is that through interactive operations combined with the GrabCut algorithm, the patterns in the blue and white porcelain images can be automatically extracted. The user only needs to select the foreground area and the background area, and the system can automatically segment and save them, which greatly improves the efficiency of blue and white porcelain pattern processing.

[0128] At the same time, the present invention ensures that the final extracted patterns are more refined through multiple image segmentation and background removal.

[0129] Specific implementation steps:

[0130] (1) Image loading and preliminary processing: Load the blue and white porcelain image and perform preliminary denoising and enhancement processing on the image to improve the details of the pattern.

[0131] (2) First GrabCut processing: Use the GrabCut algorithm to process the image for the first time and extract the preliminary foreground and background areas.

[0132] (3) Multiple corrections to the selection: Based on the preliminary processing results, the user can make multiple corrections by adjusting the selection of the rectangular box to gradually optimize the extraction of the foreground area.

[0133] (4) Second GrabCut Processing: After the user correction, a second round of GrabCut processing is performed to extract the patterns in a refined manner and remove small background parts.

[0134] (5) Final saving and display: After the pattern extraction is completed, the system provides a preview function, and the user can save the final result after confirmation. During the saving process, the user can select the path and naming rules.

[0135] The technical effect achieved is that through multi-stage GrabCut algorithm processing, the extraction of blue and white porcelain patterns is more accurate, which can effectively remove complex backgrounds and retain pattern details, and is suitable for the processing of complex patterns.

[0136] Based on the above-mentioned method, the present invention also proposes a system for extracting and classifying blue and white porcelain patterns, which is used to implement the above-mentioned method through specific unit modules. The system includes:

[0137] An image acquisition module is used to acquire a two-dimensional plane image of the blue and white porcelain decoration to obtain a source image;

[0138] A pattern separation module, used for separating the background of the source image using a GrabCut algorithm to obtain a pattern image;

[0139] An interference removal module is used to determine whether the pattern image is adhered, and if so, to remove background interference by marking the foreground and background of the pattern image;

[0140] The pattern classification module is used to send the pattern image output by the interference removal module to the pre-trained ResNet50 deep learning network model to obtain the classification corresponding to the pattern image;

[0141] A pattern storage module is used to store pattern images and their corresponding categories in a visual manner;

[0142] The ResNet50 deep learning network model takes the pattern image as input and the classification corresponding to the pattern image as output for training.

[0143] The system of the present invention first uses GrabCut to extract blue and white porcelain decorative patterns, and then uses a deep learning model to classify the extracted decorative patterns.

[0144] Specific implementation steps:

[0145] (1) Image loading and foreground extraction: After the user loads the blue and white porcelain image, the GrabCut algorithm is used to extract the foreground part of the pattern.

[0146] (2) Deep learning model classification: The extracted decorative patterns are input into a pre-trained deep learning model (e.g., ResNet50) for classification, and the model gives a prediction result based on the decorative pattern type.

[0147] (3) Result display and saving: The classification results are displayed to the user through a graphical interface, and the user can choose to save the classification results and extracted patterns.

[0148] (4) Batch classification and processing: Supports batch import of multiple blue and white porcelain images. The system automatically extracts and classifies patterns and provides batch saving and naming functions.

[0149] The technical effect achieved is that the image segmentation technology and deep learning model are combined to make the extraction and classification process of blue and white porcelain patterns more efficient and accurate. The addition of deep learning can improve the accuracy of pattern classification, and the automated processing flow greatly improves work efficiency.

[0150] The above description is only a preferred embodiment of the present invention and does not limit the present invention in any way. Any simple modification, change and equivalent structural change made to the above embodiment based on the technical essence of the present invention still falls within the protection scope of the technical solution of the present invention.

Claims

1. A method for extracting and classifying blue and white porcelain patterns, characterized in that: include: S1: Acquire a two-dimensional plane image of blue and white porcelain decoration to obtain a source image; S2: Using GrabCut algorithm to separate the background of the source image to obtain a pattern image; S3: determining whether the pattern image is adhered, and if so, removing background interference by marking the foreground and background of the pattern image; S4: Send the pattern image outputted in step S3 to a pre-trained ResNet50 deep learning network model to obtain a classification corresponding to the pattern image; S5: storing the pattern images and their corresponding categories in a visual manner; The ResNet50 deep learning network model takes the pattern image as input and the classification corresponding to the pattern image as output for training.

2. A method for extracting and classifying blue and white porcelain patterns according to claim 1, characterized in that: After step S1 and before step S2, the method further includes: The source image is smoothed by using the RTV algorithm.

3. A method for extracting and classifying blue and white porcelain patterns according to claim 2, characterized in that: The step of smoothing the source image by using the RTV algorithm includes: Performing pixel normalization processing on the source image; Calculate the relative total variation and local total variation of the source image after normalization, and determine the smoothness of each pixel point of the source image; The source image is smoothed according to the smoothness of each pixel and the set objective function.

4. A method for extracting and classifying blue and white porcelain patterns according to claim 1, characterized in that: In step S2, the GrabCut algorithm is used to separate the background of the source image to obtain a pattern image, including: Initializing the source image by stroking; According to each pixel point in the source image, an energy function of the source image is established; The source image is cut using graph cutting technology to minimize the energy function, and the cut source image is output to obtain the pattern image.

5. A method for extracting and classifying blue and white porcelain patterns according to claim 4, characterized in that: The energy function consists of a data term and a smoothing term.

6. A method for extracting and classifying blue and white porcelain patterns according to claim 1, characterized in that: In step S1, the acquisition of the source image includes: Acquire a blue and white porcelain image, wherein the blue and white porcelain image is a two-dimensional plane image; A decorative area in the blue and white porcelain image is selected as the source image according to user interaction.

7. A system for extracting and classifying blue and white porcelain patterns, characterized in that: include: An image acquisition module is used to acquire a two-dimensional plane image of the blue and white porcelain decoration to obtain a source image; A pattern separation module, used for separating the background of the source image using a GrabCut algorithm to obtain a pattern image; An interference removal module is used to determine whether the pattern image is adhered, and if so, to remove background interference by marking the foreground and background of the pattern image; The pattern classification module is used to send the pattern image output by the interference removal module to the pre-trained ResNet50 deep learning network model to obtain the classification corresponding to the pattern image; A pattern storage module is used to store pattern images and their corresponding categories in a visual manner; The ResNet50 deep learning network model takes the pattern image as input and the classification corresponding to the pattern image as output for training.

Images