Method for generating model chart and related product

Abstract

The invention discloses a method for generating a model chart and a related product. The method comprises the following steps: segmenting a human body area and a clothes area based on an original image, wherein the human body area is not covered by clothes; extracting clothing edge features, human body edge features and human body posture features based on the original image; taking the human body edge features as boundary conditions, and performing pixel information reconstruction conforming to the human body posture features on the human body region; and taking the clothing edge features as boundary conditions, and reserving pixel information of the clothing area to obtain a target model chart. Through the method disclosed by the embodiment of the invention, the boundary during pixel reconstruction can be limited by utilizing the human body edge features and the clothing edge features obtained by segmentation, and the pixel information reconstruction process is guided by utilizing the human body posture features obtained by identification, so that the risk of clothing deformation and distortion of the model chart is reduced, and the availability rate of the generated model chart is guaranteed.

Description

Technical Field

[0001] The present disclosure generally relates to the field of image processing technology. More specifically, the present disclosure relates to a method for generating a model image, an electronic device, a computer-readable storage medium, a computer program product, and an apparatus for generating a model image. Background Art

[0002] With the rapid development of computer vision and deep learning technology, the application areas of image generation and editing technology cover many aspects from artistic creation to industrial application. In e-commerce, image generation technology can be used to edit and optimize product images, such as removing backgrounds, adjusting colors, improving image quality, etc., to promote product images at the sales level.

[0003] For clothing products, model photos can intuitively show how the clothing looks on the body, and models of different ages, looks, and body shapes can show the clothing in a diversified way, thereby increasing consumers' purchasing interest. However, traditional model photo shooting involves high costs. Photographer fees, venue rental, and model fees will result in high economic costs. With the increase in the number of products and the demand for diversified display effects, economic costs such as photographer fees, venue rental, and model fees will further increase, and will also lead to longer shooting time, thereby increasing time costs.

[0004] In order to solve the high cost problem of traditional model photo shooting process, more and more companies use computer vision and deep learning technology to generate model images for clothing display. However, existing model photo generation solutions often have problems such as clothing deformation and distortion in the model photos, as well as obvious errors in the transition area of ​​the model photos, which makes the generated model photos unusable.

[0005] In view of this, there is an urgent need to provide a model image generation solution in order to improve the accuracy and reliability of the image generation process, reduce the risk of clothing deformation and distortion in the model image, and ensure the availability of the generated model image. Summary of the invention

[0006] In order to at least solve one or more of the technical problems mentioned above, the present disclosure proposes a model image generation solution in multiple aspects.

[0007] In a first aspect, the present disclosure provides a method for generating a model image, including: segmenting a human body area and a clothing area based on an original image, wherein the human body area is not covered by clothing; extracting clothing edge features, human body edge features, and human body posture features based on the original image; using the human body edge features as boundary conditions, reconstructing pixel information of the human body area that conforms to the human body posture features; and using the clothing edge features as boundary conditions, retaining the pixel information of the clothing area, and obtaining a target model image.

[0008] In a second aspect, the present disclosure provides an electronic device, comprising: a processor; and a memory storing executable program instructions, wherein when the program instructions are executed by the processor, the device implements any method according to the first aspect.

[0009] In a third aspect, the present disclosure provides a computer-readable storage medium having stored thereon a computer program code for generating a model image, and when the computer program code is executed by a processing device, any method of the first aspect is executed.

[0010] In a fourth aspect, the present disclosure provides a computer program product, comprising a computer program for generating a model image, wherein when the computer program is executed by one or more processors, the method of any one of the first aspects is implemented.

[0011] In a fifth aspect, the present disclosure provides an apparatus for generating a model image, comprising: an image input module for receiving an original image; an image segmentation module connected to the image input module for acquiring the original image from the image input module and segmenting the original image into a human body region and a clothing region based thereon; a posture extraction module connected to the image input module for acquiring the original image from the image input module and extracting human body posture features based thereon; an edge detection module connected to the image input module for acquiring the original image from the image input module and extracting clothing edge features and human body edge features based thereon; a large model generation module connected to the image input module, the image segmentation module, the posture extraction module and the edge detection module, wherein the posture extraction module and the edge detection module are control modules of the large model generation module, and the large model generation module is used to reconstruct the pixel information of the human body region provided by the image segmentation module and retain the pixel information of the clothing region under the control conditions of the posture extraction module and the edge detection module, so as to obtain a target model image; the control conditions include: taking the human body edge features and the clothing edge features as boundary conditions, and the pixel information of the reconstructed human body region conforms to the human body posture features; and an image output module connected to the large model generation module for outputting the target model image.

[0012] Through the method for generating a model image as provided above, the disclosed embodiment locates the area to be reconstructed in the original image through the extracted human body edge features, thereby defining a boundary for the pixel information reconstruction operation in the model image generation process, preventing the reserved area defined by the clothing edge features from being affected by the pixel reconstruction and causing deformation or distortion. At the same time, the method of the disclosed embodiment also guides the process of pixel information reconstruction with the extracted human body posture features, ensuring that the display posture of the clothing before and after reconstruction remains consistent, avoiding the negative impact of the authenticity of the target model image on the mismatch between the clothing form and the model posture, thereby improving the accuracy and reliability of the target model image and ensuring the availability of the target model image. BRIEF DESCRIPTION OF THE DRAWINGS

[0013] By reading the detailed description below with reference to the accompanying drawings, the above and other objects, features and advantages of the exemplary embodiments of the present disclosure will become readily understood. In the accompanying drawings, several embodiments of the present disclosure are shown in an exemplary and non-limiting manner, and the same or corresponding reference numerals represent the same or corresponding parts, wherein:

[0014] Figure 1 A schematic diagram of the generation effect of an existing model image generation solution is shown;

[0015] Figure 2 An exemplary flow chart of a method for generating a model image according to an embodiment of the present disclosure is shown;

[0016] Figure 3 A schematic diagram of edge features and a schematic diagram of human body posture features extracted by the disclosed embodiment are shown;

[0017] Figure 4 An exemplary flow chart showing a method for generating a model image according to some embodiments of the present disclosure is shown;

[0018] Figure 5 An exemplary flow chart showing a method for generating a model image according to some embodiments of the present disclosure is shown;

[0019] Figure 6 An exemplary flow chart showing a method for generating a model image according to some embodiments of the present disclosure is shown;

[0020] Figure 7 An exemplary flow chart showing a method for generating a model image according to some embodiments of the present disclosure

[0021] Figure 8 An exemplary flow chart showing a method for generating a model image according to some embodiments of the present disclosure is shown;

[0022] Fig. 9 An exemplary flow chart showing a method for generating a model image according to some embodiments of the present disclosure is shown;

[0023] Fig.10 An exemplary structural diagram showing an apparatus for generating a model image according to some embodiments of the present disclosure;

[0024] Fig.11 A schematic diagram showing a comparison of the operating effects of a large model generation module in some embodiments of the present disclosure is shown;

[0025] Fig.12 A schematic diagram showing the comparison of the operation effect of the large model generation module when the posture extraction module is used as the control module is shown;

[0026] Fig.13A schematic diagram showing the comparison of the operation effect of the large model generation module when there is no edge detection module as the control module;

[0027] Fig.14 An exemplary structural block diagram of an electronic device according to an embodiment of the present disclosure is shown. DETAILED DESCRIPTION

[0028] The following will be combined with the drawings in the embodiments of the present disclosure to clearly and completely describe the technical solutions in the embodiments of the present disclosure. Obviously, the described embodiments are part of the embodiments of the present disclosure, not all of the embodiments. Based on the embodiments in the present disclosure, all other embodiments obtained by those skilled in the art without creative work are within the scope of protection of the present disclosure.

[0029] It should be understood that the terms "include" and "comprising" used in the specification and claims of the present disclosure indicate the presence of described features, integers, steps, operations, elements and / or components, but do not exclude the presence or addition of one or more other features, integers, steps, operations, elements, components and / or collections thereof.

[0030] It should also be understood that the terms used in this disclosure are only for the purpose of describing specific embodiments and are not intended to limit the disclosure. As used in this disclosure and claims, the singular forms of "a", "an", and "the" are intended to include the plural forms unless the context clearly indicates otherwise. It should also be further understood that the term "and / or" used in this disclosure and claims refers to any combination of one or more of the associated listed items and all possible combinations, including these combinations.

[0031] As used in this specification and claims, the term "if" may be interpreted as "when" or "upon" or "in response to determining" or "in response to detecting," depending on the context. Similarly, the phrase "if it is determined" or "if [described condition or event] is detected" may be interpreted as meaning "upon determination" or "in response to determining" or "upon detection of [described condition or event]" or "in response to detecting [described condition or event]," depending on the context.

[0032] The specific implementation of the present disclosure is described in detail below with reference to the accompanying drawings.

[0033] Exemplary application scenarios

[0034] The traditional model photo shooting process will incur high shooting costs, including but not limited to: venue rental, photographer fees, model fees, etc. In addition, in order to pursue a variety of display effects, the same product may also require models of different body shapes and ages to shoot display pictures, further increasing the shooting costs. In order to reduce costs, more and more companies use computer vision and deep learning technology to generate model images for clothing displays. For example, based on the existing model pictures, the models are aged through image processing to form model pictures of different ages. For another example, a mannequin is used to try on clothing, and realistic limbs are intelligently generated through deep learning technology to obtain a virtual model picture.

[0035] However, in the existing model image generation schemes, the segmentation algorithm and the generation algorithm are completely separate and lack joint optimization. Therefore, during runtime, problems such as clothing deformation and distortion in the model image and obvious erroneous transition areas often occur in the model image, making the generated model image unusable.

[0036] For ease of understanding, Figure 1 FIG. 4 shows a schematic diagram of the generation effect of an existing model image generation solution, such as Figure 1 As shown, there is an obvious wrong transition area in the generated image on the right, and the clothing on the model is deformed and distorted compared to the original image on the left.

[0037] Exemplary Application Scenarios

[0038] In view of this, the disclosed embodiment provides a model image generation scheme, which limits the boundaries of pixel reconstruction by segmenting the human body edge features and clothing edge features, and uses the recognized human body posture features to guide the process of pixel information reconstruction, thereby preventing the occurrence of deformation problems, and ensuring that the display posture of the clothing before and after reconstruction remains consistent to prevent distortion, thereby improving the availability of the target model image.

[0039] Figure 2 An exemplary flow chart of a method 200 for generating a model image according to an embodiment of the present disclosure is shown. Figure 2 As shown, in step S201, the human body area and clothing area are segmented based on the original image. In this step, the original image is a display image of a model wearing clothing, and the display image of the model here can be a display image of a real model or a display image of a dummy model. Furthermore, the display image of the real model can be a real model image taken in a real environment, or it can be a virtual model image synthesized by artificial intelligence. The dummy model display image is a mannequin image obtained by taking a mannequin wearing clothing.

[0040] The human body region in the original image refers to the pixel region in the original image that is not covered by clothing and belongs to the model, including the torso and head region of the model.

[0041] It should be noted that the disclosed embodiment has no strict restrictions on the input and output formats of the original image, and the original image can be in png, jpg, bmp or other formats. Similarly, the disclosed embodiment has no strict restrictions on the resolution of the original image, and the method provided by the disclosed embodiment is compatible with a variety of image formats and resolutions, thereby adapting to different application scenarios.

[0042] As an example, step S201 can be performed using the SAM model (Segment Anything Model), which contains three components: an image encoder, a prompt encoder, and a mask decoder. The image encoder can map the image to a feature space to extract rich visual features, which are then encoded into a dense representation to provide support for subsequent segmentation tasks. The prompt encoder is responsible for converting user-provided prompts such as clicks, box selections, and text into an embedded form that the model can understand, so that the SAM model can flexibly respond to various forms of input prompts, thereby achieving interactive segmentation. The task of the mask decoder is to combine the outputs of the image encoder and the prompt encoder to generate the final segmentation mask. The SAM model can handle various image segmentation tasks with high flexibility and efficiency, and it has zero-sample generalization capabilities. It can segment unfamiliar objects and images without additional training, so it is very suitable for the solution of the disclosed embodiment to segment accurate human body areas and clothing areas.

[0043] As another example, a human body parsing pre-model (Human Parse) can also be used to perform step S201. The human body parsing pre-model is a model based on computer vision technology to understand the human body structure and segment different parts of the human body. The model is pre-trained on a large amount of image data and can recognize and distinguish various parts of the human body, such as the head, arms, torso, and legs. Since the human body parsing pre-model is pre-trained on a large-scale data set, it can provide powerful feature extraction and segmentation capabilities for human body parsing tasks, so it is very suitable for downstream tasks centered on people, such as the segmentation task of segmenting human body areas and clothing areas in the disclosed embodiment.

[0044] It should be further explained that the above segmentation model is an exemplary description of the embodiment of the present disclosure. In actual application, there are other segmentation models applicable to the present disclosure, which will not be repeated here.

[0045] Furthermore, due to clothing occlusion, the human body region in the original image will appear as multiple connected domains, so there is a situation where the human body region is identified as multiple regions during segmentation. Taking the SAM model as an example, the SAM model can detect the hair region, facial region, arm region and leg region from the original image, and at this time, the final human body region can be obtained by taking the union of these regions. In other words, the execution process of step S201 can be as follows: the original image is instance segmented to obtain a segmentation mask, and then the segmentation mask is processed by taking the union to obtain the human body region and the clothing region.

[0046] In step S202, clothing edge features, human body edge features and human body posture features are extracted based on the original image. In this step, human body edge features refer to the edge features of the human body region in step S201, that is, the edge features of the pixel region in the original image that is not covered by clothing and belongs to the model. For ease of understanding, Figure 3 Schematic diagram of edge features extracted by the disclosed embodiment and schematic diagram of human posture features are shown, such as Figure 3 As shown, the image on the left is the original image. The edge feature diagram in the middle can be extracted by the edge detection algorithm, and the human body posture feature diagram on the right can be extracted by the posture estimation algorithm.

[0047] based on Figure 3 As shown, it can be understood that the execution of step S202 can be divided into the following two sub-steps: using the edge detection algorithm to process the original image to extract clothing edge features and human body edge features, and using the posture estimation algorithm to extract human body key points to form human body posture features. It should be noted that there is no difference in the execution sequence of the above two sub-steps. In the actual operation process, the edge detection algorithm can be executed first, the posture estimation algorithm can be executed first, and the edge detection algorithm and the posture estimation algorithm can also be executed simultaneously.

[0048] Furthermore, human key points include both torso key points and facial key points. The torso key points include the head, neck, shoulders, elbows, wrists, hips, knees and ankles. These torso key points help to build a skeleton model of the human body, thereby estimating the posture of the human body. For example, the 16 key points of human posture annotation defined in the MPII dataset include torso key points such as right ankle, right knee, right hip, left hip, left knee, left ankle, pelvis, chest, upper neck and top of the head. Facial key points include the positions of eyes, nose, mouth and ears. These key points can be identified by a dedicated facial key point detector, such as dlib's face key point detector, which can detect 68 key points of the face, which helps to understand human expressions and head postures.

[0049] As an example, the edge detection algorithm can be implemented by edge detection models such as the Canny operator, the Sobel operator, the Roberts operator, and the Prewitt operator. The Canny operator is mainly used to extract useful structural information from an image, especially edge information. The Canny operator mainly generates clear edge features through steps such as Gaussian filtering, gradient calculation, non-maximum suppression, double threshold detection, edge tracking, and hysteresis threshold determination. The Sobel operator detects edges by calculating the first-order derivative approximation of the image grayscale function. It contains operators in the horizontal and vertical directions, and the operators in the two directions are sensitive to edges in the horizontal and vertical directions, respectively. Specifically, the Sobel operator uses two 3x3 convolution kernels to calculate the grayscale weighted difference of the central pixel neighborhood, and then extracts edge features in the image. The Roberts operator is an edge detection operator based on local difference operation, which detects edges by calculating the gradients in the horizontal and vertical directions. The Prewitt operator belongs to the first-order differential operator, which detects edges by calculating the grayscale difference of the upper, lower, left, and right neighboring points of a pixel point and reaching an extreme value at the edge.

[0050] As an example, the pose estimation algorithm can be implemented through many pose estimation models including Openpose. The technical principle of OpenPose is to use Part Affinity Fields (PAFs for short) for multi-person pose estimation. PAFs is a vector field used to represent the correlation between human body parts, thereby solving the occlusion and overlap problems in multi-person pose estimation. OpenPose extracts image features through convolutional neural networks, and combines PAFs to accurately identify the key points of multiple human bodies in the image in real time, thereby realizing pose estimation.

[0051] In step S203, pixel information of the human body region is reconstructed. In this step, the pixel information needs to comply with certain restrictions before reconstruction can be performed, wherein the boundary conditions in the restrictions are provided by the human body edge features, and the information conditions are provided by the human body posture features. In other words, the reconstruction operation shall not be performed in the area beyond the human body edge, and the reconstructed pixel information needs to conform to the human body posture features in the original image.

[0052] In step S204, the pixel information of the clothing area is retained to obtain the target model image. In this step, the retained pixel information is also subject to the boundary conditions, which are provided by the clothing edge features. In order to ensure that the clothing worn by the model before and after generation is not deformed and distorted, the pixel information belonging to the clothing area divided by the clothing edge features needs to be completely retained to ensure that the display effect of the clothing products in the target model image is consistent with the display effect of the clothing products in the original image.

[0053] As an example, step S203 and step S204 can be implemented by large-scale pre-trained models such as GPT-4, DALL-E and StableDiffusion, based on deep neural network technology, to generate the required target model graph through effective learning. GPT-4 is a Transformer-based model that can process multiple types of inputs, including text, images and audio, which makes it have strong multimodal capabilities and can better understand context, make inferences and perform complex language tasks. DALL-E is a multimodal pre-trained model, whose operation includes two stages. The first stage is the discrete variational auto-encoder (dVAE) to generate the token of the image, and the second stage is the Transformer-based generative model to mix the image and text features, convert the input description into a machine-understandable format, and then gradually build the image. Stable Diffusion is an image generation model based on deep learning, which adopts the improved MMDiT-X architecture and training method, introduces the query-key normalization technology, so as to enhance its responsiveness to prompts.

[0054] It should be noted that, in actual application, there is no strict restriction on the execution sequence of step S203 and step S204. In addition to the execution sequence shown above, step S204 can also be executed before step S203 or in parallel with step S203. No excessive restrictions are made here.

[0055] In the method shown in the embodiment of the present disclosure, the segmentation information obtained by the segmentation algorithm, edge detection algorithm and posture estimation algorithm can be applied in the generation model, thereby forming a joint optimization between image segmentation, feature extraction and image generation, ensuring the advancement and high performance of the entire image processing system, and realizing efficient and high-quality image generation.

[0056] A specific application scenario of the method shown in the embodiment of the present disclosure is to perform aging processing on the models in the original image, so as to generate clothing display pictures of models of different ages. In this application scenario, the original image can use a real model image or a virtual model image. The real model image refers to a model image obtained by taking a photo of a real person wearing clothes in a real environment, and the virtual model image refers to a clothing display picture synthesized by a virtual model synthesized based on artificial intelligence.

[0057] Figure 4 An exemplary flow chart of a method 400 for generating a model image according to some embodiments of the present disclosure is shown. Figure 4As shown, in step S401, the human body region and clothing region are segmented based on the original image. In this embodiment, the human body region refers to the pixel region belonging to the model's body that is not covered by clothing, which may include: an uncovered torso region, an uncovered facial region, and a hair region.

[0058] It should be noted that the segmentation algorithm or segmentation model used in step S401 can refer to the segmentation algorithm or segmentation model described in step S201 in the previous embodiment, and the specific execution process of step S401 can also refer to the specific execution process described in step S201 in the previous embodiment, which will not be repeated here.

[0059] In step S402, clothing edge features, human edge features and human posture features are extracted based on the original image. In this embodiment, the method for extracting edge features can refer to the description of step S202 in the above embodiment, and the method for extracting human posture features can also adopt the same extraction method as step S202 in the above embodiment.

[0060] In addition, since the original image in this embodiment is a real model image or a virtual model image, both of which have clear model facial features, such as facial expressions, when extracting human posture features, the human posture features extracted in step S402 include body posture features and facial expression features. In other words, when executing step S402, this embodiment needs to extract the key points of the trunk to construct a skeleton model of the human body, and also needs to extract the key points of the face to understand the expression and head posture of the model, so as to help the expression and posture of the model in the target model image to be consistent with the original image.

[0061] In step S403, an age feature is determined. In this embodiment, the age feature can be used to guide the reconstruction of pixel information of the human body region in the model aging processing scheme. Specifically, in some embodiments, the age feature can correspond to an age value.

[0062] In step S404, texture information is called according to the age feature. Models of different ages may have different texture information in the human body area. For example, models with a larger age value are more likely to have wrinkles and sagging skin than models with a smaller age value. This is also texture information commonly used to distinguish different age features in daily life. For another example, for some skin colors, models with a larger age value are more likely to have white hair color than models with a smaller age value. This is also texture information commonly used to distinguish different age features.

[0063] Based on this, the texture information called according to age characteristics may include: skin texture information and hair color information, wherein the skin texture information is used to form wrinkles in the target model image, and the hair color information is used to change the hair color of the model in the target model image, thereby achieving a visual effect of the model's aging.

[0064] In some embodiments, the system may pre-store a database of texture information and a table of mapping relationships between age features and texture information. After the age feature is determined in step S403, the texture information matching the age feature is screened out by table lookup and called.

[0065] In step S405, pixel information of the human body region is reconstructed according to the texture information to make it conform to the human body posture characteristics. As described in the above steps, the texture information may include skin texture information and hair color information, the human body region may include a trunk region, a facial region and a hair region, and the human body posture characteristics may include body posture characteristics and facial expression characteristics.

[0066] When executing step S405, the pixel information of the trunk area can be reconstructed according to the skin texture information to make it conform to the body posture characteristics. The pixel information of the face area can also be reconstructed according to the skin texture information to make it conform to the facial expression characteristics. The hair color information is used to reconstruct the pixel information of the hair area.

[0067] It should be noted that the reconstruction of pixel information in the three areas of the torso, face and hair also needs to comply with certain restrictions, and the edge features of the human body need to be used as boundary conditions when reconstructing pixel information.

[0068] It should be further explained that the reconstruction of the pixel information of the three regions of the torso region, the face region and the hair region can be performed simultaneously or in any execution sequence, and no excessive restrictions are imposed here. In addition, when implementing the method provided in this embodiment, it is only necessary to ensure that step S403 and step S404 are performed before step S405. In other words, step S403 and step S404 can also be performed before any of step S401 and step S402.

[0069] In step S406, the pixel information of the clothing area is retained to obtain the target model image. In this embodiment, the specific execution method of step S406 is consistent with step S204 in the above embodiment, and will not be repeated here.

[0070] The above describes a method for performing aging processing on a model in an original image. In the above method, it is necessary to first determine the aging target of the aging processing, that is, the age feature. The present disclosure supports multiple methods for determining age features, such as methods specified by the user or automatically selected by the system. Based on the two exemplary age feature determination methods described above, some embodiments of the present disclosure also provide the following two methods for generating a model image.

[0071] Figure 5 An exemplary flow chart of a method 500 for generating a model image according to some embodiments of the present disclosure is shown. Figure 5 As shown, step S501 and step S502 can be consistent with step S201 and step S202 in the above embodiment, and are intended to obtain human body region, clothing region, clothing edge features, human body edge features and human body posture features. Specifically, in step S501, human body region and clothing region are segmented based on the original image. In step S502, clothing edge features, human body edge features and human body posture features are extracted based on the original image.

[0072] In step S503, the imported data or selected instructions fed back by the user through the interactive device are obtained. The image processing system in this embodiment provides an interactive device for realizing the human-computer interaction function, through which the user can obtain the data output by the system, and can also input instructions and data to the system through the interactive device to control the image processing process.

[0073] In step S504, the age feature is determined according to the imported data or the selected instruction. As an example, the interactive device may be an interactive terminal with a screen, and a selection list containing multiple age features is displayed on the screen. The user inputs a selection instruction to the system by clicking a button corresponding to one of the age features, so that the control system calls the corresponding texture information based on the age feature pointed to by the selected instruction. As another example, the interactive device may be an interactive terminal with an input device, and the user may import a data into the system through the input device, and the imported data reflects the age feature, and the system calls the corresponding texture information based on the age feature reflected by the imported data.

[0074] After step S504 is executed, the age feature is determined, and then steps S505 to S507 can be executed in sequence to fuse the human body edge feature, clothing edge feature, human body posture feature and age feature, and process the pixel information of the human body area and the clothing area to obtain the target model map. Specifically, in step S505, the texture information is called according to the age feature. In step S506, the human body edge feature is used as a boundary condition, and the pixel information of the human body area is reconstructed according to the texture information to make it conform to the human body posture feature. In step S507, the clothing edge feature is used as a boundary condition, and the pixel information of the clothing area is retained to obtain the target model map.

[0075] It should be noted that the specific implementation method of steps S505 to S507 has been described in detail in steps S404 to S406 of the previous embodiment and will not be repeated here.

[0076] Figure 6 An exemplary flow chart of a method 600 for generating a model image according to some embodiments of the present disclosure is shown. Figure 6 As shown, step S601 and step S602 can be consistent with step S201 and step S202 in the above embodiment, aiming to obtain human body region, clothing region, clothing edge feature, human body edge feature and human body posture feature. Specifically, in step S601, human body region and clothing region are segmented based on the original image. In step S602, clothing edge feature, human body edge feature and human body posture feature are extracted based on the original image.

[0077] In step S603, the original texture information of the human body area in the original image is extracted. In some embodiments, the system may pre-store a database of texture information and a mapping table of age features and texture information. The system may filter out and call matching texture information based on the determined age features. Similarly, the system may also determine matching age features based on known texture information. Step S603 is intended to obtain the known original texture information reflected by the original image.

[0078] In step S604, the original age feature corresponding to the original texture information is determined. According to the mapping relationship pre-stored in the system, the original texture information of the human body area in the original image reflects the age feature of the model in the current model image, that is, the original age feature in the original image.

[0079] In step S605, an age feature is determined based on the original age feature. In this step, since the model in the original image needs to be aged, the age value corresponding to the age feature is greater than the age value corresponding to the original age feature. Furthermore, there may be multiple age features whose age values ​​are greater than the original age feature. In some embodiments, the system may randomly select an age feature that meets the requirements. In other embodiments, multiple or even all age features may be selected.

[0080] After step S605 is executed, the age feature is determined, and then steps S606 to S608 can be performed in sequence to fuse the human body edge feature, clothing edge feature, human body posture feature and age feature, and process the pixel information of the human body area and the clothing area to obtain the target model map. Specifically, in step S606, the texture information is called according to the age feature. In step S607, the pixel information of the human body area is reconstructed according to the texture information to make it meet the human body posture feature. In step S608, the pixel information of the clothing area is retained to obtain the target model map.

[0081] It should be noted that the specific execution method of steps S606 to S608 has been described in detail in steps S404 to S406 of the previous embodiment and will not be repeated here.

[0082] The above describes the specific implementation of the method shown in the embodiment of the present disclosure in the aging processing scenario. The following describes another specific application scenario of the method shown in the embodiment of the present disclosure. The requirement of this application scenario is to realize the anthropomorphism of the mannequin, that is, to synthesize a virtual model image with a realistic body based on the image of the mannequin wearing clothes. In this application scenario, the original image uses the mannequin image.

[0083] Figure 7 An exemplary flow chart of a method 700 for generating a model image according to some embodiments of the present disclosure is shown. Figure 7 As shown, in step S701, a human body region and a clothing region are segmented based on the original image. In step S702, clothing edge features, human body edge features and human body posture features are extracted based on the original image. Steps S701 to S702 are consistent with the contents of steps S201 to S202 in the above embodiment, and the purpose is to obtain the human body region, clothing region, clothing edge features, human body edge features and human body posture features, so as to fuse multiple features to reconstruct the pixel information of the human body region and retain the pixel information of the clothing region when subsequently generating the target model image.

[0084] It should be noted that the human body region in the mannequin image includes the uncovered human body model region, and the human body posture features in the mannequin image include body posture features.

[0085] In step S703, the target model features are determined. Mannequins are industrial production assembly line products. Unlike real models, they do not have the personalized appearance features, skin features, and expression features that are unique to real models. In order to achieve the anthropomorphism of the mannequin, it is necessary to first determine the anthropomorphism direction of the mannequin, that is, determine the target model features. As an example, the target model features may include features such as age, race, gender, and emotions to guide the appearance and expression of the virtual model in the target model image.

[0086] In step S704, skin texture information, facial expression information and facial features information are called according to the target model characteristics. As an example, skin texture information can be called according to age, and further, skin texture information can also include skin color information, and skin color information can be called according to race. Similarly, facial features information can also be called according to race and gender, and facial expression information can be called according to emotion.

[0087] In some embodiments, there is overlap between facial expression information and facial feature information, and facial expression information can affect the shape and position of facial features. Therefore, in actual application, facial expression information and facial feature information can be classified as the same information, such as facial information, and facial information can be called in combination with multiple characteristics such as race, gender, and emotion.

[0088] In step S705, the pixel information of the human model area is reconstructed according to the skin texture information, the facial expression information, the facial features information and the body posture characteristics. In this step, the skin texture information is used to guide the reconstruction of the pixel information of the human model area so that the virtual model in the target model image meets the target model characteristics. The facial expression information and the facial features information are used to guide the generation of the head of the virtual model so that it has realistic facial posture characteristics. When reconstructing the pixel information of the torso in the human model area, it is necessary to comply with the boundary conditions given by the human body edge characteristics and the restrictive conditions given by the body posture characteristics.

[0089] In step S706, the pixel information of the clothing area is retained to obtain the target model image. In this embodiment, the content of step S706 is consistent with step S204 in the above embodiment, and will not be repeated here.

[0090] Regardless of the above application scenarios, the background in the target model image can be the same as the background in the original image, or a new background can be generated. Figure 8 An exemplary flow chart of a method 800 for generating a model image according to some embodiments of the present disclosure is shown.

[0091] like Figure 8As shown, in step S801, a human body region, a clothing region and a background region are segmented based on the original image. Similar to step S201 in the foregoing embodiment, step S801 can also be performed using a SAM model or a human body analysis pre-model. In addition, since there may be a variety of decorative elements in the background region in the original image, the segmentation result classifies it into multiple regions. Therefore, referring to the processing method of the segmentation result of the human body region, after obtaining the segmentation mask, multiple regions that should belong to the background region can be merged into the background region by taking the union method.

[0092] In step S802, clothing edge features, human body edge features and human body posture features are extracted based on the original image. In this embodiment, the content of step S802 is consistent with that of step S202 in the above embodiment, and will not be repeated here.

[0093] In step S803, the clothing edge features and the human body edge features are used as boundary conditions to retain the pixel information of the background area. Since the original image can be deconstructed into three areas: the human body area, the clothing area and the background area, once the human body edge features corresponding to the human body area and the clothing edge features corresponding to the clothing area are determined, the edge features corresponding to the background area are also determined. In other words, the union of the clothing edge features and the human body edge features can be established as the edge features corresponding to the background area.

[0094] Since the present embodiment uses the background in the original image, therefore, referring to the processing method of the clothing area, step S803 uses clothing edge features and human body edge features as boundary conditions to retain the pixel information of the background area, so that the background in the target model image is consistent with the background in the original image.

[0095] In step S804, the human body edge features are used as boundary conditions to reconstruct pixel information of the human body region that conforms to the human body posture features. In this embodiment, step S804 is consistent with step S203 in the above embodiment, and will not be repeated here.

[0096] In step S805, the clothing edge features are used as boundary conditions, the pixel information of the clothing area is retained, and the target model image is obtained. In this embodiment, the content of step S805 is consistent with that of step S204 in the above embodiment, and will not be repeated here.

[0097] It should be noted that Figure 8 The method shown can be combined with the method described in any of the above embodiments, for example, Figure 4 In the embodiment of the model aging processing described above, when the pixel information of the human body area is reconstructed using the texture information, the pixel information of the clothing area and the background area is retained. Figure 7The described embodiment of the mannequin anthropomorphism processing retains the pixel information of the clothing area and the background area when reconstructing the pixel information of the human body area using the corresponding information of the target model's features.

[0098] In order to further improve the quality of the target model image, some embodiments of the present disclosure will optimize the details of the target model image after generating the target model image to obtain an optimized target model image. The method described in any of the above embodiments can add a step of detail optimization to improve the quality of the target model image.

[0099] Below Figure 2 The method shown is used as an example to introduce the specific implementation method of detail optimization processing. Fig. 9 An exemplary flow chart of a method 900 for generating a model image according to some embodiments of the present disclosure is shown. Fig. 9 As shown, steps S901 to S904 in this embodiment are Figure 2 The steps S201 to S204 are consistent with each other, and the area to be reconstructed in the original image is located by extracting the human body edge features, thereby defining the boundary for the pixel information reconstruction operation in the model image generation process. At the same time, the extracted human body posture features are used to guide the process of pixel information reconstruction, ensuring that the display posture of the clothing before and after reconstruction remains consistent.

[0100] Specifically, in step S901, a human body region and a clothing region are segmented based on the original image. In step S902, clothing edge features, human body edge features and human body posture features are extracted based on the original image. In step S903, the human body region is reconstructed with pixel information that conforms to the human body posture features. In step S904, the pixel information of the clothing region is retained to obtain a target model image.

[0101] After step S904 is completed, a target model image can be obtained, and then step S905 is performed to optimize the target model image. Specifically, in step S905, the target model image is subjected to detail optimization processing to obtain an optimized target model image. The detail optimization processing includes at least one of the following processing: edge repair, texture enhancement, illumination adjustment, and resolution enhancement.

[0102] First, the detail optimization process of edge repair is introduced. In the process of edge repair, the edge area formed by the edge features of the clothing and the edge features of the human body is repaired. This detail optimization process can repair the unnatural edges in the target model image. In the embodiment disclosed herein, the unnatural edges mainly appear in the edge area formed by the reconstructed area and the retained area. The edge area can be determined by the following steps: first, the human body area is expanded to obtain a first morphological area, then the human body area is corroded to obtain a second morphological area, and finally, the first morphological area and the second morphological area are differenced to obtain the edge area. It should be noted that the order of obtaining the first morphological area and the second morphological area can be swapped.

[0103] As an example, after obtaining the edge area, a Lama-Inpainting repainting operation can be performed on this edge area. This operation can erase the unnatural edges and make the transition more natural. Lama-Inpainting is an image restoration technology based on deep learning. It can process missing or damaged areas in the image and generate a repaired image. It should be noted that the image restoration technology described above is only an example. In actual application, edge restoration can also be achieved using image restoration technologies such as PowerPaint and BrushNet.

[0104] Next, the detail optimization process of texture enhancement is introduced. The embodiments disclosed herein mainly enhance the skin texture and clothing texture of the model, that is, enhance the texture details of the human body area and clothing area in the target model image, aiming to improve the visibility and clarity of the texture in the target model image, thereby improving the authenticity of the target model image and the clarity of the clothing.

[0105] This embodiment is applicable to a variety of texture enhancement technologies, including but not limited to: spatial domain filtering, frequency domain filtering, morphological operations, image enhancement of adaptive fractional-order operators or texture enhancement based on deep learning, etc. The above-mentioned various texture enhancement technologies can be used alone or in combination.

[0106] As an example, some embodiments of the present disclosure can achieve texture enhancement through a large model, such as using emilianJR-epiCRealism, an image generation model based on deep learning, which can generate highly realistic images, especially task scenarios for generating portraits of people, which can show high details, such as skin texture and light and shadow effects. It should be noted that when emilianJR-epiCRealism generates a target model image based on various features, it can perform a texture enhancement operation together, thereby directly outputting an optimized target model image.

[0107] The following is an introduction to the detail optimization process of illumination adjustment, which involves improving the quality of images under different illumination conditions, such as detail recovery in low-light environments. As an example, some embodiments of the present disclosure may implement illumination adjustment through emilianJR-epiCRealism, an image generation model based on deep learning. Specifically, hint words related to illumination and shadows, such as "natural shadows of sunlight", are added to the hint encoder to generate results that are as realistic as possible.

[0108] Finally, we introduce resolution enhancement, a detail optimization process, which aims to improve the clarity and details of an image. As an example, some embodiments of the present disclosure use super-resolution technology to enlarge the target model image to a specific size to improve the clarity of the target model image. Specifically, the Upscale function of the image generation model can be used, or a traditional deep learning model, such as CodeFormer, can be used.

[0109] In order to execute the method described in any of the above embodiments, some embodiments of the present disclosure provide a device for generating a model image. Fig.10 An exemplary structural diagram of an apparatus 1000 for generating a model image according to some embodiments of the present disclosure is shown.

[0110] like Fig.10 As shown, the device includes the following modules: an image input module 1010, an image segmentation module 1020, a posture extraction module 1030, an edge detection module 1040, a large model generation module 1050 and an image output module 1060.

[0111] The image input module 1010 is a data interface for receiving the original image. As an example, the image input module 1010 can be implemented as a hardware interface, and its hardware structure can be implemented as a serial interface or a parallel interface for data transmission between the system and the outside. As another example, the image input module 1010 can also be implemented as a software interface, such as an application programming interface, to enable various software applications within the system to communicate with each other.

[0112] The image segmentation module 1020 is a functional module connected to the image input module 1010. The original image enters the image segmentation module 1020 through the image input module 1010. The image segmentation module 1020 segments the human body area and the clothing area by executing the segmentation algorithm. In actual application, the image segmentation module 1020 can be a processing chip equipped with an image segmentation model. The storage circuit inside the chip is used to store the code of the image segmentation model. The processing chip runs the image segmentation model by executing the code.

[0113] The functional modules connected to the image input module 1010 also include a posture extraction module 1030 and an edge detection module 1040. The original image enters the posture extraction module 1030 and the edge detection module 1040 through the image input module 1010. The posture extraction module 1030 completes the extraction of human posture features, and the edge detection module 1040 completes the extraction of clothing edge features and human edge features. Similar to the image segmentation module 1020, in actual application, the posture extraction module 1030 can be a processing chip equipped with a posture estimation model, and the storage circuit inside the chip is used to store the code of the posture estimation model, and the processing chip runs the posture estimation model by executing the code. Similarly, the edge detection module 1040 can be a processing chip equipped with an edge detection operator, and the storage circuit inside the chip is used to store the code of the edge detection operator, and the processing chip runs the edge detection operator by executing the code.

[0114] The output ends of the four modules, image input module 1010, image segmentation module 1020, posture extraction module 1030 and edge detection module 1040, are all connected to the large model generation module 1050, which is equipped with an image generation model. Exemplarily, the large model generation module 1050 can be implemented as an integrated chip, and there is a storage circuit inside the integrated chip for storing the running code of the image generation model. Further, the chip structure formed by the connection of the posture extraction module 1030, the edge detection module 1040 and the large model generation module 1050, or the integrated circuit formed by the connection of the posture extraction module 1030, the edge detection module 1040 and the large model generation module 1050 can be regarded as a large model. From the perspective of the network structure of the large model, the posture extraction module 1030 and the edge detection module 1040 provide a sub-network that plays a control role in the large model network structure, that is, the posture extraction module 1030 and the edge detection module 1040 are control modules of the large model generation module.

[0115] Under the control conditions of the posture extraction module 1030 and the edge detection module 1040, the large model generation module 1050 reconstructs the pixel information of the human body area provided by the image segmentation module 1020 and retains the pixel information of the clothing area to obtain the target model image. Further, the control conditions are as described in the above method embodiment, which include: using the human body edge features and clothing edge features as boundary conditions, and the pixel information of the reconstructed human body area must conform to the human body posture features.

[0116] In order to more clearly show the influence of control conditions on the generation results, or in other words, to more clearly show the influence of the posture extraction module and the edge detection module on the operation results of the large model generation module, Fig.11A schematic diagram showing a comparison of the operating effects of the large model generation module in some embodiments of the present disclosure is shown. Fig.12 FIG. 4 shows a schematic diagram comparing the operation effect of the large model generation module when the posture-free extraction module is used as the control module. Fig.12 As shown in the figure, when only edge features are used for boundary control and there is a lack of control conditions for human posture features, the face of the model in the target model image will appear obviously unreal. Fig.13 FIG. 4 shows a schematic diagram comparing the operation effect of the large model generation module when there is no edge detection module as the control module. Fig.13 As shown in FIG. 1 , when there is only the control condition of the human body posture feature and the lack of edge features for boundary control, transition errors will occur in the model's hair in the target model image.

[0117] Similar to the image input module 1010, the device 1000 is also provided with an image output module 1060. In order to provide the target model image for the user, the image output module 1060 is connected to the output end of the large model generation module 1050 and implemented as a hardware interface, and its hardware structure can be implemented as a serial interface or parallel interface for data transmission between the system and the outside. Further, through this hardware interface, an interactive device can be connected to the device 1000. Further, this hardware interface can be the image input module 1010 and the image output module 1060. The interactive device inputs the original image, imports data and / or selects instructions to the device 1000 through the hardware interface as the image input module 1010, and the interactive device obtains the target model image from the device 1000 through the hardware interface as the image output module 1060.

[0118] In order to implement the method steps described in the foregoing text of this disclosure in conjunction with the accompanying drawings at the software and hardware level, the present disclosure also provides the following Fig.14 The electronic device shown. Specifically, Fig.14 An exemplary structural block diagram of an electronic device 1400 according to an embodiment of the present disclosure is shown.

[0119] like Fig.14 As shown, the electronic device 1400 disclosed herein may include a processor 1410 and a memory 1420. Specifically, executable program instructions are stored in the memory 1420. When the program instructions are executed by the processor 1410, the electronic device implements the method steps described above in conjunction with the accompanying drawings.

[0120] It is understood that in order to clearly illustrate the solution of the present disclosure and avoid confusion with the prior art, Fig.14The electronic device 1400 of the present disclosure only shows the components related to the embodiment of the present disclosure, and omits the components that may be necessary for implementing the embodiment of the present disclosure but belong to the scope of the prior art. Therefore, based on the content disclosed in the present disclosure, a person skilled in the art can clearly understand that the electronic device 1400 of the present disclosure can also include components related to the embodiment of the present disclosure. Fig.14 The constituent elements shown in are different from the common constituent elements.

[0121] In an exemplary implementation scenario, the above-mentioned processor 1410 can control the overall operation of the electronic device 1400. For example, the processor 1410 can control the operation of the electronic device 1400 by executing the program stored in the memory 1420. In terms of implementation, the processor 1410 of the present disclosure can be implemented by a central processing unit (CPU), an application processor (AP), an artificial intelligence processor chip (IPU), etc. provided in the electronic device 1400. Further, the processor 1410 of the present disclosure can also be implemented in any appropriate manner. For example, the processor 1410 can take the form of a computer-readable medium, a logic gate, a switch, an application-specific integrated circuit (ASIC), a programmable logic controller, and an embedded microcontroller, etc., such as a microprocessor or a processor and a computer-readable program code (such as software or firmware) that can be executed by the (micro) processor.

[0122] In terms of storage content, the memory 1420 can be used to store hardware of various data and instructions processed in the electronic device 1400. For example, the memory 1420 can store processed data and data to be processed in the electronic device 1400. The memory 1420 can store data sets that have been processed or to be processed by the processor 1410. In addition, the memory 1420 can store applications, drivers, etc. to be driven by the electronic device 1400. For example: the memory 1420 can store various programs related to model building and constraint setting to be executed by the processor 1410. The memory 1420 can be a DRAM, but the present disclosure is not limited thereto. In terms of type, the memory 1420 may include at least one of a volatile memory or a non-volatile memory. The non-volatile memory may include a read-only memory (ROM), a programmable ROM (PROM), an electrically programmable ROM (EPROM), an electrically erasable programmable ROM (EEPROM), a flash memory, a phase change RAM (PRAM), a magnetic RAM (MRAM), a resistive RAM (RRAM), a ferroelectric RAM (FRAM), and the like. The volatile memory may include a dynamic RAM (DRAM), a static RAM (SRAM), a synchronous DRAM (SDRAM), a PRAM, an MRAM, an RRAM, a ferroelectric RAM (FeRAM), etc. In an embodiment, the memory 1420 may include at least one of a hard disk drive (HDD), a solid state drive (SSD), a high-density flash memory (CF), a secure digital (SD) card, a micro secure digital (Micro-SD) card, a mini secure digital (Mini-SD) card, an extreme digital (xD) card, caches, or a memory stick.

[0123] In summary, the specific functions implemented by the memory 1420 and the processor 1410 of the electronic device 1400 provided in the implementation manner of this specification can be explained in comparison with the aforementioned implementation manner in this specification, and can achieve the technical effects of the aforementioned implementation manner, and will not be repeated here.

[0124] Additionally or optionally, the present disclosure may also be implemented as a non-temporary machine-readable storage medium (or computer-readable storage medium, or machine-readable storage medium, or computer program product) on which computer program instructions (or computer program, or computer instruction code) for generating a model image are stored. When the computer program instructions (or computer program, or computer instruction code) are executed by a processor of an electronic device (or electronic device, server, etc.), the processor executes part or all of the steps of the above-mentioned method according to the present disclosure.

[0125] The foregoing content can be better understood in accordance with the following terms:

[0126] Clause A1. A method for generating a model image, comprising: segmenting a human body region and a clothing region based on an original image, wherein the human body region is not covered by clothing; extracting clothing edge features, human body edge features and human body posture features based on the original image; using the human body edge features as boundary conditions, reconstructing pixel information of the human body region that conforms to the human body posture features; and using the clothing edge features as boundary conditions, retaining the pixel information of the clothing region to obtain a target model image.

[0127] Clause A2. According to the method described in Clause A1, the original image includes: a real model image, a virtual model image or a mannequin image; if the original image is a real model image or a virtual model image, the human body area includes: an uncovered torso area, an uncovered facial area and a hair area, and the human body posture features include: body posture features and facial expression features; and if the original image is a mannequin image, the human body area includes: an uncovered human body model area, and the human body posture features include: body posture features.

[0128] Clause A3. According to the method described in Clause A2, if the original image is a real model image or a virtual model image, reconstructing the pixel information of the human body area that conforms to the human body posture characteristics includes: determining age characteristics; calling texture information based on the age characteristics; and reconstructing the pixel information of the human body area based on the texture information to make it conform to the human body posture characteristics.

[0129] Clause A4. According to the method described in Clause A3, the texture information includes: skin texture information and hair color information, wherein reconstructing the pixel information of the human body area based on the texture information to make it conform to the human body posture characteristics includes: reconstructing the pixel information of the torso area based on the skin texture information to make it conform to the body posture characteristics; reconstructing the pixel information of the facial area based on the skin texture information to make it conform to the facial expression characteristics; and reconstructing the pixel information of the hair area based on the hair color information.

[0130] Clause A5. According to the method described in Clause A2, if the original image is a mannequin image, reconstructing pixel information of the human body area that conforms to human body posture characteristics includes: determining target model characteristics; calling skin texture information, facial expression information and facial features information according to the target model characteristics; and reconstructing pixel information of the human body model area according to skin texture information, facial expression information, facial features information and body posture characteristics.

[0131] Clause A6. The method according to Clause A1, wherein segmenting the human body area and the clothing area based on the original image includes: segmenting the human body area, the clothing area and the background area based on the original image; wherein after extracting clothing edge features, human body edge features and human body posture features based on the original image, the method for generating a model image also includes: using clothing edge features and human body edge features as boundary conditions to retain pixel information of the background area.

[0132] Clause A7. The method according to Clause A1, wherein after obtaining the target model image, the method for generating the model image further comprises: performing detail optimization processing on the target model image to obtain an optimized target model image; wherein the detail optimization processing comprises one or more of the following: edge repair, texture enhancement, lighting adjustment and resolution enhancement.

[0133] Clause A8. According to the method described in Clause A7, the detail optimization processing includes: edge repair, wherein the detail optimization processing of the target model image includes: repairing the edge area formed by the edge features of the clothing and the edge features of the human body.

[0134] Clause A9. According to the method described in Clause A7, the detail optimization processing includes: texture enhancement, wherein the detail optimization processing of the target model image includes: enhancing the texture details of the human body area and clothing area in the target model image.

[0135] Clause A10. According to the method described in Clause A8, before repairing the edge area formed by the edge features of the clothing and the edge features of the human body, the method for generating a model image also includes: performing expansion processing and corrosion processing on the human body area respectively to obtain a first morphological area and a second morphological area; and performing difference processing on the first morphological area and the second morphological area to obtain an edge area.

[0136] Clause A11. The method according to clause A1, wherein segmenting the human body region and the clothing region based on the original image comprises: performing instance segmentation on the original image to obtain a segmentation mask; and performing a union process on the segmentation mask to obtain the human body region and the clothing region.

[0137] Clause A12. The method according to Clause A1, wherein the extraction of clothing edge features, human body edge features and human body posture features based on the original image comprises: processing the original image using an edge detection algorithm to extract clothing edge features and human body edge features; and extracting human body key points using a posture estimation algorithm to form human body posture features.

[0138] Clause A13. The method according to clause A3, wherein determining the age characteristics comprises: obtaining imported data or selected instructions fed back by the user through an interactive device; and determining the age characteristics based on the imported data or the selected instructions.

[0139] Clause A14. The method according to Clause A3, wherein determining the age feature comprises: extracting original texture information of the human body area in the original image; determining the corresponding original age feature based on the original texture information; and determining the age feature based on the original age feature, wherein the age value corresponding to the age feature is greater than the age value corresponding to the original age feature.

[0140] Item A15. An electronic device, comprising: a processor; and a memory storing executable program instructions, which, when the program instructions are executed by the processor, enables the device to implement the method of any one of Items A1 to 14.

[0141] Clause A16. A computer-readable storage medium having stored thereon a computer program code for generating a model image, wherein when the computer program code is executed by a processing device, the method of any one of clauses A1 to 14 is executed.

[0142] Clause A17. A computer program product, comprising a computer program for generating a model image, wherein when the computer program is executed by one or more processors, the method of any one of clauses A1 to 14 is implemented.

[0143] Clause A18. A device for generating a model image, comprising: an image input module for receiving an original image; an image segmentation module connected to the image input module, for obtaining the original image from the image input module and segmenting the original image into a human body area and a clothing area based on the original image; a posture extraction module connected to the image input module, for obtaining the original image from the image input module and extracting human body posture features based on the original image; an edge detection module connected to the image input module, for obtaining the original image from the image input module and extracting clothing edge features and human body edge features based on the original image; a large model generation module connected to the image input module, the image segmentation module, the posture extraction module and the edge detection module, and the posture extraction module and the edge detection module are control modules of the large model generation module, and the large model generation module is used to reconstruct the pixel information of the human body area provided by the image segmentation module under the control conditions of the posture extraction module and the edge detection module, and retain the pixel information of the clothing area to obtain a target model image; the control conditions include: using human body edge features and clothing edge features as boundary conditions, and the pixel information of the reconstructed human body area conforms to the human body posture features; and an image output module connected to the large model generation module, for outputting the target model image.

[0144] Although multiple embodiments of the present disclosure have been shown and described herein, it will be apparent to those skilled in the art that such embodiments are provided by way of example only. Those skilled in the art may think of many changes, modifications, and alternatives without departing from the thought and spirit of the present disclosure. It should be understood that in the process of practicing the present disclosure, various alternatives to the embodiments of the present disclosure described herein may be adopted. The attached claims are intended to define the scope of protection of the present disclosure, and therefore cover equivalents or alternatives within the scope of these claims.

[0145] The collection and acquisition of various data in this disclosure complies with relevant laws and regulations and is authorized by the data provider. Any organization or individual that needs to obtain external data must obtain authorization in accordance with the law and ensure data security. It is prohibited to illegally collect, use, process, or transmit unauthorized or unprotected data, or to illegally buy, sell, provide, or disclose unauthorized or unprotected data.

Claims

1. A method for generating a model graph, characterized in that: include: Segmenting a human body region and a clothing region based on the original image, wherein the human body region is not covered by the clothing; Extracting clothing edge features, human body edge features and human body posture features based on the original image; Reconstructing pixel information of the human body region that conforms to the human body posture characteristics using the human body edge characteristics as boundary conditions; and The clothing edge features are used as boundary conditions, the pixel information of the clothing area is retained, and a target model image is obtained.

2. The method according to claim 1, characterized in that The original image includes: a real model image, a virtual model image or a mannequin image; If the original image is a real model image or a virtual model image, the human body region includes: an uncovered torso region, an uncovered facial region and a hair region, and the human body posture features include: body posture features and facial expression features; and If the original image is a mannequin image, the human body region includes: an uncovered human body model region, and the human body posture features include: body posture features.

3. The method according to claim 2, characterized in that If the original image is a real model image or a virtual model image, reconstructing pixel information of the human body region that conforms to the human body posture characteristics includes: Determine age characteristics; calling texture information according to the age feature; and The pixel information of the human body region is reconstructed according to the texture information to make it conform to the human body posture characteristics.

4. The method according to claim 3, characterized in that The texture information includes: skin texture information and hair color information, wherein reconstructing the pixel information of the human body region according to the texture information so as to conform to the human body posture characteristics includes: Reconstructing pixel information of the trunk area according to the skin texture information so as to make it conform to the body posture characteristics; Reconstructing pixel information of the facial area according to the skin texture information to make it conform to the facial expression characteristics; and The pixel information of the hair area is reconstructed according to the hair color information.

5. The method according to claim 2, characterized in that: If the original image is a mannequin image, reconstructing pixel information of the human body region that conforms to the human body posture characteristics includes: Determine target model characteristics; Invoking skin texture information, facial expression information and facial features information according to the target model characteristics; and The pixel information of the human body model area is reconstructed according to the skin texture information, the facial expression information, the facial features information and the body posture characteristics.

6. The method according to claim 1, characterized in that The segmenting of the human body region and the clothing region based on the original image includes: segmenting the human body region, the clothing region and the background region based on the original image; After extracting clothing edge features, human body edge features and human body posture features based on the original image, the method further comprises: using the clothing edge features and the human body edge features as boundary conditions to retain pixel information of the background area.

7. The method according to claim 1, characterized in that After obtaining the target model image, the method further includes: Performing detail optimization processing on the target model image to obtain an optimized target model image; The detail optimization processing includes one or more of the following: edge repair, texture enhancement, lighting adjustment and resolution enhancement.

8. The method according to claim 7, characterized in that The detail optimization processing includes: edge repair, wherein the detail optimization processing of the target model image includes: The edge area formed by the edge features of the clothing and the edge features of the human body is repaired.

9. The method according to claim 7, characterized in that: The detail optimization processing includes: texture enhancement, wherein the detail optimization processing of the target model image includes: The texture details of the human body area and the clothing area in the target model image are enhanced.

10. The method according to claim 8, characterized in that Before repairing the edge area formed by the edge features of the clothing and the edge features of the human body, the method further comprises: Performing expansion processing and corrosion processing on the human body region respectively to obtain a first morphological region and a second morphological region; and The first morphological region and the second morphological region are subjected to difference processing to obtain the edge region.

11. The method according to claim 1, characterized in that: The human body area and clothing area are segmented based on the original image: Performing instance segmentation on the original image to obtain a segmentation mask; and The segmentation masks are subjected to a union process to obtain the human body region and the clothing region.

12. The method according to claim 1, characterized in that The extraction of clothing edge features, human body edge features and human body posture features based on the original image includes: Processing the original image using an edge detection algorithm to extract the clothing edge features and the human body edge features; and A posture estimation algorithm is used to extract key points of the human body to form the human body posture features.

13. The method according to claim 3, characterized in that The age characteristics include: Obtaining the input data or selected instructions fed back by the user through the interactive device; and The age characteristic is determined based on the imported data or the selected instruction.

14. The method according to claim 3, characterized in that The age characteristics include: Extracting original texture information of a human body region in the original image; Determining the original age feature corresponding to the original texture information; and The age feature is determined based on the original age feature, and the age value corresponding to the age feature is greater than the age value corresponding to the original age feature.

15. An electronic device, characterized in that: include: processor; as well as A memory storing executable program instructions, which, when executed by the processor, enables the device to implement the method according to any one of claims 1 to 14.

16. A computer program product comprising a computer program for generating a model diagram, characterized in that When the computer program is executed by one or more processors, the method according to any one of claims 1 to 14 is implemented.

Images