Clothing color changing method and multi-mode large model
Through the segmentation and color processing of clothing pictures, combined with the optimization and repair of multimodal large models, the existing clothing color modification technology is solved, and efficient and accurate clothing color modification is achieved, reducing manual participation and data Mark the requirements.
Patent Information
- Application Number
- CN202411390323.6
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-09-30
- Publication Date
- 2025-01-17
AI Technical Summary
The existing clothing color change technology takes a long time and has poor color change effect, especially the automatic color change scheme based on deep learning has problems such as inaccurate color and unstable generation effects.
A clothing color change method is adopted to segment the original clothing pictures input by the user, generate accurate clothing area masks, statistical color histograms for color removal, and combine target colors to mix the layer to extract highlights, shadows and textures. Finally, the multimodal model is optimized and repaired to generate the target color clothing picture.
The automated clothing color change process is realized, which reduces the need for manual participation and data labeling, improves the color change efficiency and the authenticity of the effect, and ensures the accuracy of the color and the naturalness of the color change effect.
Smart Images
Figure CN119323613A_ABST
Abstract
Description
Technical Field
[0001] The present invention belongs to the technical field of image rendering and machine vision, and in particular relates to a clothing color changing method and a multi-modal large model. Background Art
[0002] Clothing color changing technology is a 2D image rendering technology for ready-made clothing. This technology, while retaining the texture, fabric, pattern and design details of the original clothing, changes the partial or overall color of the clothing in a targeted manner, achieving real-time preview of clothing color matching. It can assist clothing designers in quickly realizing their ideas during the design stage, greatly accelerating the design process.
[0003] There are two main methods for clothing color change. The first is a manual color change solution based on Photoshop, but it requires human participation, has high learning costs, and is time-consuming. The second is an automatic color change solution based on deep learning, but there are problems with inaccurate colors and unstable generation effects, and it requires a large amount of labeled data, which is labor-intensive and time-consuming. Summary of the invention
[0004] The present invention provides a clothing color changing method and a multimodal large model, aiming to solve the problem that the clothing color changing technology in the prior art is time-consuming and has poor color changing effect.
[0005] In order to solve the above technical problems, the technical solution adopted by the present invention is:
[0006] In a first aspect, the present invention provides a method for changing the color of clothing, comprising the following steps:
[0007] S100, receiving an original clothing picture input by a user, segmenting the original clothing picture, obtaining masks of different clothing areas and returning the masks to the user, and the user selecting a mask of a clothing area that needs to be changed in color from the masks of different clothing areas as the mask to be changed in color;
[0008] S200, calculating a color histogram of the clothing area to be changed in the original clothing image according to the mask to be changed, and decolorizing the color histogram to obtain a decolorized image;
[0009] S300, receiving a target color input by a user; layer-mixing the target color with the decolorization image to obtain a color-changing image;
[0010] S400, extracting the highlight area, shadow area and texture on the clothing area to be changed in the original clothing picture, and merging the highlight area, shadow area and texture into the color-changed picture to obtain the color-changed clothing picture;
[0011] S500: Optimizing and repairing the color-changed clothing image to obtain a target color clothing image.
[0012] In a further solution, in step S100, the original clothing image is segmented using a objects segmentation model or a color separation model to obtain masks of different clothing areas.
[0013] Based on the above scheme, the use of the object segmentation model or the color separation model can accurately segment different areas of clothing in the original clothing image, thereby generating clear masks of different clothing areas, greatly improving the processing accuracy and automation level, and saving manpower.
[0014] In a further embodiment, when the color separation model is used to segment the original clothing image,
[0015] The color separation model segments the original clothing image to obtain masks of different clothing areas, including:
[0016] S110, converting the RGB format in the original clothing image into a LAB format;
[0017] S120, setting a main color, using a clustering algorithm to obtain the main color of the original clothing image, and obtaining pixels whose distance from the main color is less than a set threshold among all colors of the original clothing image, and using the pixels as a main color mask;
[0018] S130, inputting the main color mask into a color separation model trained by a UNet network, and the color separation model generates masks of different clothing areas according to the main color mask.
[0019] Based on the above scheme, using the color separation model to segment the original clothing image can effectively solve the problem of poor segmentation effect for hollow patterns, improve the segmentation quality, and thus obtain a more accurate mask.
[0020] In a further solution, in step S200, the step of obtaining the decolorized image includes:
[0021] S210, converting the original clothing image in RGB format into a grayscale image with a mean value of 127, and calculating a frequency histogram of each pixel value in the grayscale image to obtain the color histogram;
[0022] S220, fitting the data point set in the color histogram using a Gaussian function to obtain a mean and a variance;
[0023] S230, mapping the data point set to new pixels through a mapping formula, completing decolorization, and obtaining the decolorized image in grayscale;
[0024] S240, converting the grayscale decolorization image into a decolorization image in RGB format.
[0025] Based on the above scheme, the use of the Gaussian function can effectively smooth noise and process abnormal data points. Through fitting, the color distribution characteristics of the original clothing image can be extracted, thereby better achieving decolorization.
[0026] In a further solution, in step S230, the mapping formula is:
[0027]
[0028] Among them, the B i represents the mean, the C i represents variance; new Represents the pixel value after decolorization, the X old Represents the pixel value of the original clothing image.
[0029] In a further solution, in step S300, the target color is mixed with the decolorization image in layers by using a superimposed coating color mixing method, and the formula is:
[0030] If A≤128, then
[0031] If A>128, then
[0032] Among them, A represents the decolorization image; B represents the target color; and C represents the color change image.
[0033] In a further solution, in the step S400, the method for extracting the highlight area and the shadow area on the clothing area to be color-changed in the original clothing image is: statistically analyzing the pixel histogram of the decolorized image; in the pixel histogram, taking the secondary peak with a pixel value higher than a set value as the highlight area, and taking the secondary peak with a pixel value lower than the set value as the shadow area;
[0034] The method for extracting the texture of the clothing area to be changed in the original clothing picture is: sharpening the original clothing picture to obtain a sharpened picture, blurring the original clothing picture to obtain a blurred picture; and subtracting the sharpened picture from the blurred picture to obtain the texture.
[0035] Based on the above scheme, the highlight area and the shadow area are defined by the secondary peak of the pixel value distribution, which can not only quickly distinguish the light and dark parts in the original clothing image with high efficiency; but also more accurately retain the clothing details and avoid the loss of details caused by excessive highlights or darkness.
[0036] Sharpening the original clothing image can increase the edges and details on the clothing, making the texture of the clothing surface clearer; blurring the original clothing image can remove large areas of color blocks and low-frequency information; the texture obtained by subtracting the sharpened image from the blurred image can effectively highlight the details in the clothing.
[0037] A further solution: in the step S500, the color-changed clothing picture is optimized and repaired by a multimodal large model; the multimodal large model generates the target color clothing picture according to the color-changed clothing picture.
[0038] Based on the above scheme, the authenticity of the target color clothing picture can be improved by optimizing and repairing the color-changed clothing picture.
[0039] In a second aspect, the present invention provides a multimodal large model for clothing color change, which is used to perform a clothing color change method as described in any of the first aspects; the multimodal large model includes:
[0040] The segmentation module receives the original clothing picture input by the user, segments the original clothing picture, obtains masks of different clothing areas and returns them to the user, and the user selects the mask of the clothing area that needs to be changed in color from the masks of different clothing areas as the mask to be changed in color;
[0041] The color mapping module calculates the color histogram of the clothing area to be changed in the original clothing picture according to the mask to be changed, and decolorizes the color histogram to obtain a decolorized image; receives a target color input by a user; and mixes the target color with the decolorized image in layers to obtain a changed color image;
[0042] A light and texture effect restoration module extracts the highlight area, shadow area and texture on the clothing area to be changed in the original clothing image, and merges the highlight area, shadow area and texture into the color-changed image to obtain the color-changed clothing image;
[0043] The secondary restoration module optimizes and restores the color-changed clothing image to obtain a target color clothing image.
[0044] A further solution: before optimizing and repairing the color-changed clothing image through the secondary repair module, the multimodal large model is first trained, and the training steps include:
[0045] Step 1: collecting pure color clothing pictures, masks of the pure color clothing pictures and colors of the pure color clothing pictures, and inputting them into the multimodal large model;
[0046] Step 2: The color is layer-mixed with the decolorized image of the pure color clothing image by the color mapping module, and the highlight area, shadow area and texture on the pure color clothing image are integrated by the light and shadow and texture effect restoration module to obtain a color-fitting clothing image;
[0047] Step 3: the multimodal large model regenerates the solid color clothing picture according to the color fitting clothing picture; and obtains the trained multimodal large model.
[0048] Based on the above scheme, the multimodal large model is trained to learn the color distribution and characteristics in the solid color clothing pictures, so that the trained multimodal large model can generate higher quality and more realistic color-changed clothing pictures, thereby improving the color-changing effect.
[0049] The beneficial effects of the present invention are:
[0050] 1. The method for changing the color of clothing in the present invention is completely completed automatically by the multimodal large model, without the need for manual participation or large amounts of data annotation, which is highly efficient, saves labor, and reduces costs. In addition, the present invention can achieve directional color change of a part or the entire area of clothing, and the color after color change is accurate. The color-changed clothing picture is optimized and repaired by the secondary repair module, further improving the authenticity of the color change effect.
[0051] 2. The present invention can achieve a natural, delicate and highly personalized color change effect by finely segmenting the original clothing image, extracting the color of the clothing, retaining the highlight shadow and texture, and optimizing and repairing the clothing after color change, while greatly improving the user's controllability and satisfaction with the clothing color change. BRIEF DESCRIPTION OF THE DRAWINGS
[0052] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings required for use in the embodiments are briefly introduced below. It should be understood that the following drawings only show certain embodiments of the present invention and therefore should not be regarded as limiting the scope. For ordinary technicians in this field, other relevant drawings can be obtained based on these drawings without paying creative work.
[0053] Figure 1 It is a schematic diagram of the steps of a method for changing the color of clothing in the present invention;
[0054] Figure 2 It is a schematic diagram of the steps of segmenting the original clothing picture by the color separation model in the present invention;
[0055] Figure 3 is a schematic diagram of the step of obtaining the decolorized image in step S200 of the present invention;
[0056] Figure 4is a schematic diagram of an original clothing picture in a specific case of the present invention;
[0057] Figure 5 is a schematic diagram of a top mask of an original clothing image in a specific example of the present invention;
[0058] Figure 6 It is a schematic diagram of a top after the color is changed in a specific case of the present invention. DETAILED DESCRIPTION
[0059] The following will be combined with the accompanying drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not used to limit the present invention. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative work belong to the protection scope of the present invention.
[0060] Embodiment 1:
[0061] like Figure 1 As shown, this embodiment provides a method for changing the color of clothing, comprising the following steps:
[0062] S100, receiving an original clothing picture input by a user, segmenting the original clothing picture, obtaining masks of different clothing areas and returning the masks to the user, and the user selecting a mask of a clothing area that needs to be changed in color from the masks of different clothing areas as the mask to be changed in color;
[0063] S200, calculating a color histogram of the clothing area to be changed in the original clothing image according to the mask to be changed, and decolorizing the color histogram to obtain a decolorized image;
[0064] S300, receiving a target color input by a user; layer-mixing the target color with the decolorization image to obtain a color-changing image;
[0065] S400, extracting the highlight area, shadow area and texture on the clothing area to be changed in the original clothing picture, and merging the highlight area, shadow area and texture into the color-changed picture to obtain the color-changed clothing picture;
[0066] S500: Optimizing and repairing the color-changed clothing image to obtain a target color clothing image.
[0067] A more specific example of step S100 in the above solution is: the mask is a binary image, the segmented area is marked as white (or 1), and the other areas are marked as black (or 0).
[0068] In step S100, the masks of different clothing areas refer to masks generated for different clothing areas or parts of the original clothing image. For example, if the original clothing image is a T-shirt, after segmenting the T-shirt, a mask of the sleeve part, a mask of the collar part, or a mask of the whole T-shirt can be obtained for the user to choose. For example, if the user wants to change the color of the collar of the T-shirt, the mask of the collar part can be selected as the mask to be changed.
[0069] In the step S100, the original clothing image is segmented using a objects segmentation model or a color separation model to obtain masks of different clothing areas.
[0070] The Segment Anything Model (SAM) is an open source segmentation model. When the SAM is used to segment the original clothing image, the steps include: receiving the original clothing image; the SAM recognizes the original clothing image and segments the clothing parts in the image; the result of the segmentation by the SAM is output in the form of a mask; the mask formed by the segmentation is post-processed (for example: smoothing edges, filling holes, etc.), and finally obtaining masks of different clothing areas.
[0071] When the color separation model is used to segment the original clothing image, Figure 2 As shown, the color separation model segments the original clothing image to obtain masks of different clothing areas, including:
[0072] S110, converting the RGB format in the original clothing image into a LAB format;
[0073] S120, setting a main color, using a clustering algorithm to obtain the main color of the original clothing image, and obtaining pixels whose distance from the main color is less than a set threshold among all colors of the original clothing image, and using the pixels as a main color mask; wherein the set threshold is pre-set by the designer;
[0074] S130, inputting the main color mask into a color separation model trained by a UNet network, and the color separation model generates masks of different clothing areas according to the main color mask.
[0075] A more specific example of the above scheme is as follows: in step S110, the RGB format in the original clothing image is obtained by changing the three color channels of red, green and blue and superimposing them with each other to obtain various colors; wherein R represents red, G represents green, and B represents blue. The LAB format in the original clothing image is a spatial color that is more in line with human visual perception; wherein L represents the brightness channel, A represents the green-red channel, and B represents the blue-yellow channel.
[0076] In step S120, the step of using a clustering algorithm (KMeans) to obtain the main color of the original clothing image includes: loading the original clothing image in LAB format and preprocessing (the preprocessing may include scaling and adjusting the image size); converting the original clothing image into a one-dimensional array for clustering, where each pixel is used as a sample and its LAB value is used as a feature; clustering the pixels using a clustering algorithm (KMeans) to find out the main color of the original clothing image.
[0077] A more specific example of step S200 in the above solution is: in step S200, if Figure 3 As shown, the step of obtaining the decolorized image includes:
[0078] S210, converting the original clothing image in RGB format into a grayscale image with a mean value of 127, and calculating a frequency histogram of each pixel value in the grayscale image to obtain the color histogram;
[0079] S220, fitting the data point set in the color histogram using a Gaussian function to obtain a mean and a variance;
[0080] S230, mapping the data point set to new pixels through a mapping formula, completing decolorization, and obtaining the decolorized image in grayscale;
[0081] S240, converting the grayscale decolorization image into a decolorization image in RGB format.
[0082] A more specific example of the above solution is: in step S220, the Gaussian function is:
[0083]
[0084] Wherein, the x represents an independent variable, which refers to the color value or the value of the color channel in the color histogram; the G i (x) represents the Gaussian function value fitting the color histogram; i Indicates amplitude; i represents the mean; i Represents variance.
[0085] In step S230, the mapping formula is:
[0086]
[0087] Among them, the B i represents the mean, the C i represents variance; new Represents the pixel value after decolorization, the X old Represents the pixel value of the original clothing image.
[0088] A more specific example of step S300 in the above solution is: in step S300, the target color and the decolorization map are mixed in layers by using a superimposed coating color mixing method, and the formula is:
[0089] If A≤≤128, then
[0090] If A>128, then
[0091] Among them, A represents the decolorization image; B represents the target color; and C represents the color change image.
[0092] A more specific example of step S400 in the above scheme is: in step S400, the method of extracting the highlight area and the shadow area on the clothing area to be changed in the original clothing image is: statistically calculating the pixel histogram of the decolorized image; in the pixel histogram, the secondary peak with a pixel value higher than a set value is taken as the highlight area, and the secondary peak with a pixel value lower than the set value is taken as the shadow area; wherein the set value is set by the designer.
[0093] In addition, the method for extracting the texture on the clothing area to be changed in the original clothing picture is: sharpening the original clothing picture to obtain a sharpened picture, blurring the original clothing picture to obtain a blurred picture; subtracting the sharpened picture from the blurred picture to obtain the texture.
[0094] A more specific example of step S500 in the above scheme is: in step S500, the color-changed clothing image is optimized and repaired through a multimodal large model; the multimodal large model generates the target color clothing image based on the color-changed clothing image.
[0095] Embodiment 2:
[0096] This embodiment provides a multimodal large model for clothing color change, which is used to execute a clothing color change method as described in Embodiment 1; the multimodal large model includes:
[0097] The segmentation module receives the original clothing picture input by the user, segments the original clothing picture, obtains masks of different clothing areas and returns them to the user, and the user selects the mask of the clothing area that needs to be changed in color from the masks of different clothing areas as the mask to be changed in color;
[0098] The color mapping module calculates the color histogram of the clothing area to be changed in the original clothing picture according to the mask to be changed, and decolorizes the color histogram to obtain a decolorized image; receives a target color input by a user; and mixes the target color with the decolorized image in layers to obtain a changed color image;
[0099] A light and texture effect restoration module extracts the highlight area, shadow area and texture on the clothing area to be changed in the original clothing image, and merges the highlight area, shadow area and texture into the color-changed image to obtain the color-changed clothing image;
[0100] The secondary restoration module optimizes and restores the color-changed clothing image to obtain a target color clothing image.
[0101] A more specific example of the above solution is: before optimizing and repairing the color-changed clothing image through the secondary repair module, the multimodal large model (stable diffusion 1.5) is first trained, and the training steps include:
[0102] Step 1: collect pure color clothing pictures (e.g., red short-sleeved shirts), the masks of the pure color clothing pictures (masks of short-sleeved shirts), and the colors of the pure color clothing pictures (red), and input them into the multimodal large model;
[0103] Step 2: The color (red) is layer-mixed with the decolorized image of the pure color clothing picture (red short-sleeved shirt) by the color mapping module, and the highlight area, shadow area and texture on the pure color clothing picture (red short-sleeved shirt) are fused by the light and shadow and texture effect restoration module to obtain a color-fitted clothing picture (i.e., the red is fitted to the decolorized red short-sleeved shirt);
[0104] Step 3: The multimodal large model regenerates the solid color clothing image (red short-sleeved shirt) according to the color-matched clothing image; and obtains the trained multimodal large model.
[0105] It should be noted that the red short-sleeved shirt in the above step of training the multimodal large model is only one example of the solid color clothing picture, which is only used to facilitate understanding of the steps of training the multimodal large model. In addition to the red short-sleeved shirt, the solid color clothing picture can also be a white dress, black trousers, etc.
[0106] The following is a further explanation of the multimodal large model for clothing color change through a specific case:
[0107] Case content: Through the multimodal large model Figure 4 of( Figure 4 The color of the top in the method (including an off-white top and off-white pants) is changed to green; the steps include:
[0108] Step 1: Figure 4 (off-white top and off-white pants) are input into the multimodal large model;
[0109] Step 2: The segmentation module Figure 4 Segmentation is performed to obtain the top mask and pants mask, etc., and the mask is returned to the user. The user selects the top mask (such as Figure 5 (shown) is the mask to be changed in color.
[0110] Step 3: The color mapping module is based on Figure 5 right Figure 4 The middle and upper garment is decolored to obtain a decolored image; the green color (green being the target color) input by the user is received, and the green color is layer-mixed with the decolored image to obtain a color-changing image (green upper garment).
[0111] Step 4: The light and texture effect repair module is Figure 4 The highlight area, shadow area and texture of the top are extracted, and the highlight area, shadow area and texture are fused to the color-changing image to obtain a color-changing clothing image (a green top fused with highlights, shadows and textures).
[0112] Step 5: The secondary repair module optimizes and repairs the color-changed clothing image to obtain a target color clothing image (such as Figure 6 The optimized green top is shown, but the color of the pants has not changed and remains off-white).
[0113] The present invention is not limited to the above-mentioned optional implementation modes. Anyone can derive other various forms of products under the inspiration of the present invention. However, no matter what changes are made in the shape or structure, all technical solutions that fall within the scope defined by the claims of the present invention fall within the protection scope of the present invention.
Claims
1. A method for changing the color of clothing, characterized in that: The following steps are involved: S100, receiving an original clothing picture input by a user, segmenting the original clothing picture, obtaining masks of different clothing areas and returning the masks to the user, and the user selecting a mask of a clothing area that needs to be changed in color from the masks of different clothing areas as the mask to be changed in color; S200, calculating a color histogram of the clothing area to be changed in the original clothing image according to the mask to be changed, and decolorizing the color histogram to obtain a decolorized image; S300, receiving a target color input by a user; layer-mixing the target color with the decolorization image to obtain a color-changing image; S400, extracting the highlight area, shadow area and texture on the clothing area to be changed in the original clothing picture, and merging the highlight area, shadow area and texture into the color-changed picture to obtain the color-changed clothing picture; S500: Optimizing and repairing the color-changed clothing image to obtain a target color clothing image.
2. A method for changing the color of clothing according to claim 1, characterized in that: In the step S100, the original clothing image is segmented using a objects segmentation model or a color separation model to obtain masks of different clothing areas.
3. A method for changing the color of clothing according to claim 2, characterized in that: When the color separation model is used to segment the original clothing image, The color separation model segments the original clothing image to obtain masks of different clothing areas, including: S110, converting the RGB format in the original clothing image into a LAB format; S120, setting a main color, using a clustering algorithm to obtain the main color of the original clothing image, and obtaining pixels whose distance from the main color is less than a set threshold among all colors of the original clothing image, and using the pixels as a main color mask; S130, inputting the main color mask into a color separation model trained by a UNet network, and the color separation model generates masks of different clothing areas according to the main color mask.
4. A method for changing the color of clothing according to claim 1, characterized in that: In step S200, the step of obtaining the decolorized image includes: S210, converting the original clothing image in RGB format into a grayscale image with a mean value of 127, and calculating a frequency histogram of each pixel value in the grayscale image to obtain the color histogram; S220, fitting the data point set in the color histogram using a Gaussian function to obtain a mean and a variance; S230, mapping the data point set to new pixels through a mapping formula, completing decolorization, and obtaining the decolorized image in grayscale; S240, converting the grayscale decolorization image into a decolorization image in RGB format.
5. A method for changing the color of clothing according to claim 4, characterized in that: In step S230, the mapping formula is: Among them, the B i represents the mean, the C i represents variance; new Represents the pixel value after decolorization, the X old Represents the pixel value of the original clothing image.
6. A method for changing the color of clothing according to claim 1, characterized in that: In the step S300, the target color is layer-mixed with the decolorization image by using a superimposed coating color mixing method, and the formula is: If A≤≤128, then If A>128, then Among them, A represents the decolorization image; B represents the target color; and C represents the color change image.
7. A method for changing the color of clothing according to claim 1, characterized in that: In the step S400, the method for extracting the highlight area and the shadow area on the clothing area to be color-changed in the original clothing image is: statistically analyzing the pixel histogram of the decolorized image; in the pixel histogram, taking the secondary peak with a pixel value higher than a set value as the highlight area, and taking the secondary peak with a pixel value lower than the set value as the shadow area; The method for extracting the texture of the clothing area to be changed in the original clothing picture is: sharpening the original clothing picture to obtain a sharpened picture, blurring the original clothing picture to obtain a blurred picture; and subtracting the sharpened picture from the blurred picture to obtain the texture.
8. A method for changing the color of clothing according to claim 1, characterized in that: In the step S500, the color-changed clothing picture is optimized and repaired by using a multimodal large model; the multimodal large model generates the target color clothing picture according to the color-changed clothing picture.
9. A multi-modal large model for clothing color change, characterized in that: Used to perform a clothing color changing method as described in any one of claims 1 to 8; the multimodal large model includes: The segmentation module receives the original clothing picture input by the user, segments the original clothing picture, obtains masks of different clothing areas and returns them to the user, and the user selects the mask of the clothing area that needs to be changed in color from the masks of different clothing areas as the mask to be changed in color; The color mapping module calculates the color histogram of the clothing area to be changed in the original clothing picture according to the mask to be changed, and decolorizes the color histogram to obtain a decolorized image; receives a target color input by a user; and mixes the target color with the decolorized image in layers to obtain a changed color image; A light and texture effect restoration module extracts the highlight area, shadow area and texture on the clothing area to be changed in the original clothing image, and merges the highlight area, shadow area and texture into the color-changed image to obtain the color-changed clothing image; The secondary restoration module optimizes and restores the color-changed clothing image to obtain a target color clothing image.
10. The multi-modal large model for clothing color change according to claim 9, characterized in that: Before optimizing and repairing the color-changed clothing image through the secondary repair module, the multimodal large model is first trained, and the training steps include: Step 1: collecting pure color clothing pictures, masks of the pure color clothing pictures and colors of the pure color clothing pictures, and inputting them into the multimodal large model; Step 2: The color is layer-mixed with the decolorized image of the pure color clothing image by the color mapping module, and the highlight area, shadow area and texture on the pure color clothing image are integrated by the light and shadow and texture effect restoration module to obtain a color-fitting clothing image; Step 3: the multimodal large model regenerates the solid color clothing picture according to the color fitting clothing picture; and obtains the trained multimodal large model.