Image style transfer method and transfer system based on cutting

Abstract

The present invention provides an image style transfer method based on cutting and a transfer system thereof, belonging to the technical field of image processing. The present invention realizes accurate style conversion and retention of real-time image content through fine area division and style matching, multi-dimensional quantitative analysis of image features, and adaptive transfer optimization strategy, can dynamically evaluate the transfer effect, and implement iterative improvement for non-standard areas until the expected style standard is reached, thereby ensuring that the style transfer work not only shows rich artistic expression but also is highly faithful to the details of the original image while maintaining high efficiency, and realizes a delicate, harmonious and high-quality image style transfer effect.

Description

Technical Field

[0001] The present invention relates to the field of image processing technology, and in particular to a method for image style transfer based on image cutting and a transfer system thereof. Background Art

[0002] In the field of digital media and art, image style transfer has long attracted much attention as a creative technology. It aims to combine the content of one image with the style of another image to create a unique and attractive visual effect. However, existing style transfer technologies face multiple challenges, especially when processing real-time images. How to achieve high fidelity and naturalness of style transfer while ensuring processing speed has become a key issue that needs to be solved. Specifically, the main problems include:

[0003] 1. Traditional style transfer methods often adopt global processing strategies, which makes it difficult to accurately control the application of styles in different areas of the image, and easily lead to excessive mixing of styles or unnatural boundary effects;

[0004] 2. During the conversion process, how to retain the original image content details while fully integrating the target style features to avoid the loss of content information or the bluntness of style expression is a major challenge;

[0005] 3. There is a lack of effective mechanisms to automatically evaluate the quality of migrated images, especially in terms of style consistency and artistic expression, and to perform adaptive optimization based on the evaluation results.

[0006] Therefore, it is necessary to provide an image style transfer method based on cutting and a transfer system thereof to solve the above technical problems. Summary of the invention

[0007] In order to solve the above technical problems, the present invention provides an image style transfer method based on cutting and its transfer system to achieve accurate style conversion and retention of real-time image content, ensure that the style transfer work not only shows rich artistic expression, but also is highly faithful to the details of the original image, and achieves a delicate, harmonious and high-quality image style transformation effect.

[0008] The image style transfer method based on cut-out provided by the present invention comprises the following steps:

[0009] S1: Divide the image to be processed into regions based on image features to obtain at least one sub-region, and match style templates for all sub-regions based on a preset regional style matching model to obtain a preliminary style transfer map corresponding to the sub-region;

[0010] S2: Extract image features from the preliminary style-transferred image and map the image features into a multi-dimensional space to obtain transfer quantitative data;

[0011] S3: based on a preset migration section, determining the location of the migration quantification data belonging to the migration section, and executing a corresponding migration optimization strategy according to the determined location, wherein the migration section includes a migration standard-reaching section and a migration standard-unreaching section;

[0012] If the preliminary style transfer map is determined to belong to the migration standard section, the preliminary style transfer map is fused to generate a final style transfer map;

[0013] If the preliminary style transfer map is determined to belong to a section where the transfer fails to meet the standard, step S4 is executed;

[0014] S4: superimposing the preliminary style transfer map to the corresponding sub-region in the image to be processed, and re-executing steps S1, S2 and S3 until the preliminary style transfer map in step S3 is determined to belong to the migration standard segment.

[0015] Preferably, the step S1 specifically includes the following steps:

[0016] S101: Identify image features of an image to be processed, and divide the image to be processed into a plurality of sub-regions containing local features using an image segmentation algorithm;

[0017] S102: Matching the image features of each sub-region with a corresponding style template based on a pre-trained regional style matching model;

[0018] S103: Generate a corresponding preliminary style transfer map based on the style template of each sub-region.

[0019] Preferably, in step S102, the step of constructing the regional style matching model includes:

[0020] S1021: Acquire a historical image dataset of the same type as the image to be processed, and based on the sub-regions of the historical image dataset and the corresponding style templates, mark a mapping relationship between the style template and the extracted sub-region features;

[0021] S1022: Using the mapping relationship between the marked style template and the extracted sub-region features as training data, training to obtain a regional style matching model.

[0022] Preferably, step S2 specifically includes the following steps:

[0023] S201: extracting image features from the preliminary style transfer map using a feature extraction algorithm, wherein the extracted image features include color distribution, texture details, and shape contours;

[0024] S202: Perform feature mapping on all image features to form at least one feature vector;

[0025] S203: quantizing all the generated feature vectors to obtain migration quantization data.

[0026] Preferably, step S3 specifically includes the following steps:

[0027] S301: Preset two migration sections, including a migration section that meets the standards and a migration section that does not meet the standards;

[0028] S302: Determine the migration section to which the migration quantization data belongs;

[0029] S303: If it is determined that the preliminary style transfer map belongs to the migration standard section, a fusion algorithm is used to generate a final style transfer map;

[0030] S304: If it is determined that the preliminary style transfer map belongs to a section where the transfer fails to meet the standard, step S4 is executed.

[0031] Preferably, step S4 specifically includes the following steps:

[0032] S401: superimposing the preliminary style transfer map onto a corresponding sub-region in the image to be processed by an alpha blending method;

[0033] S402: re-execute steps S1, S2 and S3 for the sub-regions corresponding to the image to be processed, until all preliminary style transfer maps corresponding to the sub-regions in the image to be processed are determined to belong to the migration-qualified segment.

[0034] An image style transfer system based on cut-out, the transfer system comprising:

[0035] A division and matching module, used to divide the image to be processed into regions based on image features to obtain at least one sub-region, and match style templates for all sub-regions based on a preset regional style matching model to obtain a preliminary style transfer map corresponding to the sub-region;

[0036] The feature extraction and quantification module is used to extract image features from the preliminary style transfer image and map the image features into a multi-dimensional space to obtain transfer quantification data;

[0037] A judgment execution module is used to judge the position of the migration quantization data belonging to the migration section based on the preset migration section, and execute the corresponding migration optimization strategy according to the judged position. In this embodiment, the migration section includes a migration standard section and a migration standard section. If the preliminary style transfer map is determined to belong to the migration standard section, the final style transfer map is generated by fusion. If the preliminary style transfer map is determined to belong to the migration standard section, the feedback execution module is run;

[0038] A feedback execution module is used to superimpose the preliminary style transfer map to the corresponding sub-region in the image to be processed, and re-run the division and matching module, the feature extraction and quantization module and the judgment execution module until the preliminary style transfer map is determined to belong to the migration standard segment through the judgment execution module.

[0039] Compared with the related art, the image style transfer method based on cut-out and the transfer system thereof provided by the present invention have the following beneficial effects:

[0040] The present invention realizes accurate style conversion and retention of real-time image content through fine region division and style matching, multi-dimensional quantitative analysis of image features, and adaptive migration optimization strategy. It can dynamically evaluate the migration effect and implement iterative improvement for substandard areas until the expected style standard is met. Thus, while maintaining high efficiency, it ensures that the style migration works not only show rich artistic expression but also are highly faithful to the details of the original image, achieving a delicate, harmonious and high-quality image style transformation effect. BRIEF DESCRIPTION OF THE DRAWINGS

[0041] Figure 1 A flow chart of the image style transfer method based on cutting provided by the present invention;

[0042] Figure 2 This is a module structure diagram of the image style transfer system based on cutting provided by the present invention. DETAILED DESCRIPTION

[0043] The present invention will be further described below in conjunction with the accompanying drawings and implementation modes.

[0044] Embodiment 1

[0045] This embodiment provides an image style transfer method based on cutting pictures. Figure 1 As shown, the migration method includes the following steps:

[0046] S1: Divide the image to be processed into regions based on image features to obtain at least one sub-region, and match style templates for all sub-regions based on a preset regional style matching model to obtain a preliminary style transfer map corresponding to the sub-region.

[0047] In this embodiment, based on key image features, the content of the image to be processed is analyzed, divided into multiple sub-regions, and each sub-region is matched with a preset style template. The fine partitioning helps to improve the naturalness and artistic effect of the overall style transfer.

[0048] S2: Extract image features from the preliminary style-transferred image and map the image features into a multi-dimensional space to obtain transfer quantitative data.

[0049] In this embodiment, image features such as color distribution, texture details, and shape contours are extracted from the preliminary style transfer image, and the PCA linear dimensionality reduction method is used to perform feature mapping on the image features such as color distribution, texture details, and shape contours to form multiple feature vectors. Then, the line quantization lq technology is used to quantize all the formed feature vectors to obtain the transfer quantization data.

[0050] Step S2 can quantitatively evaluate the degree and quality of style transfer, so that subsequent steps can decide whether further optimization is needed based on numerical standards rather than subjective judgment. Through quantification, it is easy to objectively understand and manipulate the process and results of style transfer.

[0051] S3: Based on the preset migration segment, determine the position of the migration quantization data to which the migration segment belongs, and execute the corresponding migration optimization strategy according to the determined position. In this embodiment, the migration segment includes a migration standard segment and a migration standard segment. If the preliminary style transfer map is determined to belong to the migration standard segment, the final style transfer map is generated by fusion. If the preliminary style transfer map is determined to belong to the migration standard segment, step S4 is executed.

[0052] In this embodiment, a migration segment is set in advance, and the migration quantification data of the color distribution, texture details, and shape contour features of each sub-region are judged separately, and whether they belong to the migration standard segment or the migration standard segment that does not meet the migration standard. When it is determined that the migration quantification data of the color distribution, texture details, and shape contour features of all sub-regions are determined to belong to the migration standard segment, the final style transfer map is generated by fusion. When it is determined that the migration quantification data of the color distribution, texture details, and shape contour features of any sub-region are determined to belong to the migration standard segment that does not meet the migration standard, step S4 is executed.

[0053] By setting different migration sections, you can more finely control and optimize the migration effect.

[0054] S4: superimposing the preliminary style transfer map to the corresponding sub-region in the image to be processed, and re-executing steps S1, S2 and S3 until the preliminary style transfer map is determined to belong to the migration standard segment in step S3.

[0055] In this embodiment, the preliminary style transfer map is re-superimposed on the corresponding position of the original image, and then the entire process from S1 to S3 is repeated to re-optimize the migration until the color distribution, texture details, and shape contour features of all sub-regions are determined to belong to the migration-qualified segment.

[0056] In the specific implementation process, the step S1 specifically includes the following steps:

[0057] S101: Identify image features of an image to be processed, and use an image segmentation algorithm to divide the image to be processed into a plurality of sub-regions representing local features.

[0058] In this embodiment, convolutional neural network (CNN) technology is used to identify image features such as color distribution, texture details, and shape contours of the image to be processed, and edge-based segmentation and region-based segmentation image segmentation algorithms are used to divide the image to be processed into multiple sub-regions representing local features.

[0059] For example, if the image to be processed is a night scene of a city, the image segmentation algorithm may divide the image into multiple sub-regions such as the sky, buildings, streets, vehicles, pedestrians, etc., where each sub-region has unique image features: the dark blue image features of the sky sub-region, the contour and texture features of the building sub-region, and the light detail features of the street sub-region.

[0060] S102: Match the image features of each sub-region with a corresponding style template based on a pre-trained regional style matching model.

[0061] S103: Generate a preliminary style transfer map for each sub-region using the matching corresponding style template.

[0062] In this embodiment, based on the image features of color distribution, texture details and shape contours, a style template matching each sub-region is identified. After the corresponding features are identified, the image features of color distribution, texture details and shape contours on the style template are applied to the image of the identified corresponding sub-region to generate a preliminary style transfer map for each sub-region.

[0063] For example, after style transfer, the sky sub-region becomes an image with artistic dark blue tones and cloud textures; while the building sub-region becomes more prominent in outline and texture details.

[0064] In the specific implementation process, in step S102, the step of constructing the regional style matching model includes:

[0065] S1021: Acquire a historical image dataset of the same type as the image to be processed, and based on the sub-regions of the historical image dataset and the corresponding style templates, mark a mapping relationship between the style template and the extracted sub-region features;

[0066] S1022: Using the mapping relationship between the marked style template and the extracted sub-region features as training data, training to obtain a regional style matching model.

[0067] In this embodiment, a deep learning architecture is selected, and the mapping relationship between the marked style template and the extracted sub-region features is used as training data to train the deep learning architecture to obtain a regional style matching model.

[0068] In the specific implementation process, the step S2 specifically includes the following steps:

[0069] S201: extracting image features from the preliminary style transfer map using a feature extraction algorithm, wherein the extracted image features specifically include color distribution, texture details, and shape contour image features.

[0070] In this embodiment, the SIFT algorithm is selected to detect key points at different scales of the preliminary style transfer map, and a descriptor is generated for each key point, wherein the descriptor specifically includes color distribution, texture details, and shape contour image features of the key point.

[0071] S202: Mapping the extracted image features into a multi-dimensional space to form at least one feature vector.

[0072] In this embodiment, the PCA linear dimensionality reduction method is used to map the extracted image features from the original space to the multi-dimensional feature space to form a feature vector of the color distribution, texture details and shape contour image features of each sub-region.

[0073] S203: quantizing all the generated feature vectors to obtain migration quantization data.

[0074] In this embodiment, the line quantization lq technology is used to quantize the feature vectors of the color distribution, texture details and shape contour image features of each sub-region. In this embodiment, the step of quantizing the feature vectors specifically includes setting multiple quantization points in the feature space, then calculating the distance from each feature vector to each quantization point, and quantizing the feature vector to the quantization point closest to it, to obtain migration quantization data.

[0075] In the specific implementation process, the step S3 specifically includes the following steps:

[0076] S301: Preset two migration sections, including a migration standard-compliant section and a migration standard-uncompliant section.

[0077] Specifically, in the multidimensional feature space, two areas are pre-set, including a migration standard section and a migration non-standard section. In this embodiment, the multidimensional feature space used is a two-dimensional feature space, the migration standard section is a circular area, and the migration non-standard section is all areas outside the circular area.

[0078] S302: Compare the position of the migration quantization data in the multidimensional space with the position of the migration segment to determine the segment to which it belongs.

[0079] In this embodiment, the position of the migration quantization data in the multidimensional space is compared with the position of the migration segment. In this embodiment, the migration quantization data of the color distribution, texture details and shape contour features of each sub-region are respectively represented as a point, and then it is determined whether this point is in the migration standard segment (i.e., within the circular area). If it is within the circular area, the preliminary style transfer map is considered to be up to standard; if it is not within the circular area, it is considered to be unsatisfactory.

[0080] S303: If it is determined that the preliminary style transfer map belongs to the migration standard section, a fusion algorithm is used to generate a final style transfer map.

[0081] In this embodiment, if the preliminary style transfer map is determined to belong to the migration standard segment, a fusion algorithm of color balance and detail enhancement is used to ensure that the final style transfer map maintains the target style while retaining the details and clarity of the original image.

[0082] S304: If it is determined that the preliminary style transfer map belongs to a section where the transfer fails to meet the standard, step S4 is executed.

[0083] In the specific implementation process, the step S4 specifically includes the following steps:

[0084] S401: superimposing the preliminary style transfer map onto a corresponding sub-region in the image to be processed.

[0085] In this embodiment, it is determined which sub-regions in the image to be processed need to be re-style transferred, and two images are superimposed together through an alpha blending method for subsequent reprocessing of the sub-regions.

[0086] S402: re-execute steps S1, S2 and S3 for the sub-regions corresponding to the image to be processed, until all preliminary style transfer maps corresponding to the sub-regions in the image to be processed are determined to belong to the migration-qualified segment.

[0087] Exemplarily, if the preliminary migration map of the Van Gogh style is superimposed on the architectural part of the urban landscape photo, the feature extraction, style conversion and content reconstruction steps in the style migration algorithm are re-executed to further optimize the migration effect, and then the style migration effect of the architectural part is evaluated to see whether it meets the preset migration standard (that is, if the preliminary style migration map is determined to belong to the migration standard section). If it does not meet the standard, the iterative processing of the non-standard part continues; if all parts meet the standard, the entire iterative process is terminated, and a urban landscape photo with Van Gogh-style buildings is obtained.

[0088] Embodiment 2

[0089] This embodiment provides an image style transfer system based on cut-picture. Figure 2 As shown, the migration system includes:

[0090] The division and matching module 100 is used to divide the image to be processed into regions based on image features to obtain at least one sub-region, and match style templates for all sub-regions based on a preset regional style matching model to obtain a preliminary style transfer map corresponding to the sub-region.

[0091] In this embodiment, the division and matching module is used to analyze the content of the image to be processed based on the image features of color distribution, texture details and shape contour characteristics, divide it into multiple sub-regions, and match each sub-region with a preset style template. The fine partitioning helps to improve the naturalness and artistic effect of the overall style transfer.

[0092] The feature extraction and quantification module 200 is used to extract image features from the image of the preliminary style transfer, and map the image features into a multi-dimensional space to obtain transfer quantification data.

[0093] In this embodiment, the feature extraction and quantization module is used to extract image features such as color distribution, texture details, shape outlines, etc. from the preliminary style transfer image, and use the PCA linear dimensionality reduction method to feature map the image features such as color distribution, texture details, shape outlines, etc. to form multiple feature vectors, and then use the linear quantization lq technology to quantize all the formed feature vectors to obtain migration quantization data.

[0094] The feature extraction and quantification module 300 can quantitatively evaluate the degree and quality of style transfer, so that subsequent steps can decide whether further optimization is needed based on numerical standards rather than subjective judgment. Through quantification, it is easy to objectively understand and manipulate the process and results of style transfer.

[0095] The judgment execution module 400 is used to judge the position of the migration quantization data belonging to the migration segment based on the preset migration segment, and execute the corresponding migration optimization strategy according to the judged position. In this embodiment, the migration segment includes a migration standard segment and a migration non-standard segment. If the preliminary style transfer map is determined to belong to the migration standard segment, the final style transfer map is generated by fusion. If the preliminary style transfer map is determined to belong to the migration non-standard segment, the feedback execution module is run.

[0096] In this embodiment, the judgment execution module is used to pre-set the migration segment, and at the same time, respectively judge whether the migration quantification data of the color distribution, texture details and shape contour features of each sub-region is determined to belong to the migration standard segment or the migration unqualified segment. When it is judged that the migration quantification data of the color distribution, texture details and shape contour features of all sub-regions are determined to belong to the migration standard segment, the final style transfer map is generated by fusion. When it is judged that the migration quantification data of the color distribution, texture details and shape contour features of any sub-region is determined to belong to the migration unqualified segment, the feedback execution module is run.

[0097] A feedback execution module is used to superimpose the preliminary style transfer map to the corresponding sub-region in the image to be processed, and re-run the division and matching module, the feature extraction and quantization module and the judgment execution module until the preliminary style transfer map is determined to belong to the migration standard segment through the judgment execution module.

[0098] In this embodiment, the feedback execution module is used to overlay the preliminary style transfer map back to the corresponding position of the original image, and then repeatedly run the division and matching module, the feature extraction and quantization module and the judgment execution module to re-optimize the migration until the color distribution, texture details and shape contour features of all sub-areas are determined to belong to the migration-qualified segment.

[0099] The image style transfer method based on cutting and its transfer system provided in this embodiment realizes accurate style conversion and retention of real-time image content through fine region division and style matching, multi-dimensional quantitative analysis of image features, and adaptive transfer optimization strategy. It can dynamically evaluate the transfer effect and implement iterative improvement for the substandard areas until the expected style standard is reached. Thus, while maintaining high efficiency, it ensures that the style transfer works not only show rich artistic expression but also are highly faithful to the details of the original image, achieving a delicate, harmonious and high-quality image style transfer effect.

[0100] The present application is described with reference to flowcharts and / or block diagrams of methods, devices (systems) and computer program products according to embodiments of the present application. It should be understood that each process and / or box in the flowchart and / or block diagram, as well as the combination of processes and / or boxes in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor or other programmable data processing device to generate a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 A process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0101] Those skilled in the art can understand that all or part of the steps in the various methods of the above embodiments can be completed by instructing related hardware through a program, and the program can be stored in a computer-readable storage medium, and the storage medium includes a read-only memory (ROM), a random access memory (RAM), a programmable read-only memory (PROM), an erasable programmable read-only memory (EPROM), a one-time programmable read-only memory (OTPROM), an electronically erasable rewritable read-only memory (EEPROM), a compact disc (CD-ROM) or other optical disc storage, magnetic disk storage, magnetic tape storage, or any other computer-readable medium that can be used to carry or store data.

[0102] It should also be noted that the terms "include", "comprises" or any other variations thereof are intended to cover non-exclusive inclusion, so that a process, method, commodity or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, commodity or device. In the absence of more restrictions, the elements defined by the sentence "comprises a ..." do not exclude the existence of other identical elements in the process, method, commodity or device including the elements.

Claims

1. The image style transfer method based on cutting is characterized by: The migration method comprises the following steps: S1: Divide the image to be processed into regions based on image features to obtain at least one sub-region, and match style templates for all sub-regions based on a preset regional style matching model to obtain a preliminary style transfer map corresponding to the sub-region; S2: Extract image features from the preliminary style-transferred image and map the image features into a multi-dimensional space to obtain transfer quantitative data; S3: based on a preset migration section, determining the location of the migration quantification data belonging to the migration section, and executing a corresponding migration optimization strategy according to the determined location, wherein the migration section includes a migration standard-reaching section and a migration standard-unreaching section; If the preliminary style transfer map is determined to belong to the migration standard section, the preliminary style transfer map is fused to generate a final style transfer map; If the preliminary style transfer map is determined to belong to a section where the transfer fails to meet the standard, step S4 is executed; S4: superimposing the preliminary style transfer map to the corresponding sub-region in the image to be processed, and re-executing steps S1, S2 and S3 until the preliminary style transfer map in step S3 is determined to belong to the migration standard segment.

2. The image style transfer method based on cutting according to claim 1, characterized in that: Step S1 specifically includes the following steps: S101: Identify image features of an image to be processed, and divide the image to be processed into a plurality of sub-regions containing local features using an image segmentation algorithm; S102: Matching the image features of each sub-region with a corresponding style template based on a pre-trained regional style matching model; S103: Generate a corresponding preliminary style transfer map based on the style template of each sub-region.

3. The image style transfer method based on cutting according to claim 2, characterized in that: In step S102, the steps of constructing the regional style matching model include: S1021: Acquire a historical image dataset of the same type as the image to be processed, and based on the sub-regions of the historical image dataset and the corresponding style templates, mark the mapping relationship between the style template and the extracted sub-region features; S1022: Using the mapping relationship between the marked style template and the extracted sub-region features as training data, the deep learning architecture is trained to obtain a regional style matching model.

4. The image style transfer method based on cutting according to claim 3, characterized in that: Step S2 specifically includes the following steps: S201: extracting image features from the preliminary style transfer map using a feature extraction algorithm, wherein the extracted image features include color distribution, texture details, and shape contours; S202: Perform feature mapping on all image features to form at least one feature vector; S203: quantizing all the generated feature vectors to obtain migration quantization data.

5. The image style transfer method based on cut-out according to claim 4, characterized in that: Step S3 specifically includes the following steps: S301: Preset two migration sections, including a migration section that meets the standards and a migration section that does not meet the standards; S302: Determine the migration section to which the migration quantization data belongs; S303: If it is determined that the preliminary style transfer map belongs to the migration standard section, a fusion algorithm is used to generate a final style transfer map; S304: If it is determined that the preliminary style transfer map belongs to a section where the transfer fails to meet the standard, step S4 is executed.

6. The image style transfer method based on cut-out according to claim 5, characterized in that: Step S4 specifically includes the following steps: S401: superimposing the preliminary style transfer map onto a corresponding sub-region in the image to be processed by an alpha blending method; S402: re-execute steps S1, S2 and S3 for the sub-regions corresponding to the image to be processed, until all preliminary style transfer maps corresponding to the sub-regions in the image to be processed are determined to belong to the migration-qualified segment.

7. A system for image style transfer based on cut-out, applied to the method for image style transfer based on cut-out according to any one of claims 1 to 6, characterized in that: The migration system comprises: A division and matching module, used to divide the image to be processed into regions based on image features to obtain at least one sub-region, and match style templates for all sub-regions based on a preset regional style matching model to obtain a preliminary style transfer map corresponding to the sub-region; The feature extraction and quantification module is used to extract image features from the preliminary style transfer image and map the image features into a multi-dimensional space to obtain transfer quantification data; A judgment execution module is used to judge the position of the migration quantization data belonging to the migration section based on a preset migration section, and execute a corresponding migration optimization strategy according to the judged position, wherein the migration section includes a migration standard section and a migration standard section. If the preliminary style migration map is determined to belong to the migration standard section, the final style migration map is generated by fusion. If the preliminary style migration map is determined to belong to the migration standard section, the feedback execution module is run; A feedback execution module is used to superimpose the preliminary style transfer map to the corresponding sub-region in the image to be processed, and re-run the division and matching module, the feature extraction and quantization module and the judgment execution module until the preliminary style transfer map is determined to belong to the migration standard segment through the judgment execution module.

Images