A method and apparatus for generating texture maps

Abstract

This application discloses a method and apparatus for generating texture maps, relating to the field of computer vision technology. The method includes: acquiring an initial face image of a target person; performing illumination normalization processing on the initial face image based on preset illumination information to obtain a first face image; identifying the highlight regions of the first face image and optimizing the highlight regions to obtain a second face image; and mapping the second face image onto a 3D head model of the target person to generate a texture map corresponding to the target person. In this application, by standardizing illumination information under uniform illumination conditions, uniform shadow distribution is achieved; by identifying and optimizing highlight regions, abnormal highlights are suppressed, achieving precise optimization of shadow and highlight regions in the 2D face image, avoiding the influence of irregular shadow and highlight regions on the rendering effect, thereby improving the rendering effect of the 3D face model.

Description

Technical Field

[0001] This application relates to the field of computer vision technology, and in particular to a method and apparatus for generating texture maps. Background Technology

[0002] With the continuous development of computer vision and graphics technologies, the application of generating 3D face models based on 2D face images is becoming increasingly widespread in fields such as virtual reality, digital human reconstruction, visual effects, and medical modeling. Among these, texture mapping, as a core step in the visualization of 3D face models, directly affects the realism and rendering effect of the final model.

[0003] However, under natural lighting conditions, facial images often produce local shadows and highlights due to differences in the direction and intensity of the light source or environmental reflections; and local highlights may also be caused by the secretion of oil on the surface of the facial skin or direct strong light.

[0004] In current technology, the pixel colors of the original two-dimensional face image are generally directly mapped onto the surface of the corresponding three-dimensional face model. The resulting texture map retains irregular shadow and highlight areas, resulting in poor rendering of the final three-dimensional face model. Summary of the Invention

[0005] To address the aforementioned issues, this application provides a method and apparatus for generating texture maps, enabling precise optimization of shadow and highlight areas in two-dimensional face images, avoiding the impact of irregular shadow and highlight areas on rendering effects, thereby improving the rendering effect of the corresponding three-dimensional face model.

[0006] The embodiments of this application disclose the following technical solutions:

[0007] In a first aspect, embodiments of this application disclose a method for generating texture maps, including:

[0008] Obtain the initial facial image of the target person;

[0009] Based on preset illumination information, the initial face image is subjected to illumination normalization processing to obtain a first face image; wherein, the preset illumination information is the illumination information corresponding to a pre-set standardized uniform illumination condition;

[0010] Identify the highlighted areas of the first face image and optimize the highlighted areas to obtain the second face image;

[0011] The second face image is mapped onto the 3D head model of the target person to generate a texture map corresponding to the target person.

[0012] In one possible implementation, identifying the highlighted region of the first face image includes:

[0013] Based on the saturation distribution corresponding to the first face image, identify the first bright area of ​​the first face image;

[0014] Based on the RGB channel difference corresponding to the first face image, identify the second highlight area of ​​the first face image;

[0015] Based on the first highlighted area and the second highlighted area, the highlighted area of ​​the first face image is generated.

[0016] In one possible implementation, identifying the first highlighted region of the first face image based on the saturation distribution corresponding to the first face image includes:

[0017] Based on the first face image, generate the corresponding saturation channel map;

[0018] Histogram analysis is performed on the saturation channel map to determine the saturation threshold;

[0019] Based on the saturation threshold, pixels with saturation values ​​less than the saturation threshold in the saturation channel image are filtered out to obtain the first filtering region;

[0020] An erosion and dilation operation is performed on the first selected area to obtain the first highlighted area of ​​the first face image.

[0021] In one possible implementation, identifying the second highlighted region of the first face image based on the RGB channel difference corresponding to the first face image includes:

[0022] Based on the RGB channel difference corresponding to the first face image, pixels whose RGB channel difference is less than a preset RGB channel difference threshold are selected to obtain a second filtering area;

[0023] An erosion and dilation operation is performed on the second selected area to obtain the second highlighted area of ​​the first face image.

[0024] In one possible implementation, generating the highlight region of the first face image based on the first highlight region and the second highlight region includes:

[0025] Erosion and dilation operations are performed on the first and second highlighted areas to obtain the highlighted areas of the first face image.

[0026] In one possible implementation, after performing illumination normalization processing on the initial face image based on preset illumination information to obtain the first face image, the method further includes:

[0027] Identify the face region and background region in the first face image;

[0028] Based on the face region, the background region is inpainted to obtain the inpainted background region;

[0029] Based on the repaired background region and the face region, a repaired first face image is generated;

[0030] The process of identifying the highlighted areas of the first face image and optimizing those areas to obtain the second face image includes:

[0031] The highlighted areas of the repaired first face image are identified, and the highlighted areas are optimized to obtain the second face image.

[0032] In one possible implementation, identifying the face region and background region in the first face image includes:

[0033] Based on the first face image, a segmentation mask map of the face region is generated using a face segmentation algorithm;

[0034] Based on the segmentation mask map of the face region, identify the face region in the first face image;

[0035] Inversely select the segmentation mask image of the face region to generate a segmentation mask image of the background region;

[0036] Based on the segmentation mask map of the background region, the background region in the first face image is identified.

[0037] In one possible implementation, the method further includes:

[0038] Facial landmark detection is performed on the second face image to obtain the coordinates of the facial landmarks; wherein, the coordinates of the facial landmarks include: the coordinates of the corner of the mouth landmarks and the coordinates of the cheek landmarks;

[0039] Based on the coordinates of the key points at the corners of the mouth, the shadow area at the corners of the mouth in the second face image is determined;

[0040] Based on the coordinates of the cheek key points, determine the cheek replacement area in the second face image that has the same area as the shadow area at the corner of the mouth;

[0041] Based on the cheek replacement area, pixel replacement processing is performed on the corner of the mouth shadow area to obtain the replaced second face image.

[0042] In one possible implementation, the step of performing pixel replacement processing on the corner-of-mouth shadow region based on the cheek replacement region to obtain a replaced second face image includes:

[0043] Map the pixel values ​​of the cheek replacement area to the corner of the mouth shadow area;

[0044] The Poisson fusion algorithm is used to smooth the shadow area at the corner of the mouth after mapping, so as to obtain the second face image after replacement.

[0045] Secondly, embodiments of this application disclose a texture mapping generation apparatus, comprising:

[0046] The image acquisition module is used to acquire the initial facial image of the target person;

[0047] The illumination processing module is used to perform illumination normalization processing on the initial face image based on preset illumination information to obtain a first face image; wherein, the preset illumination information is the illumination information corresponding to a pre-set standardized uniform illumination condition;

[0048] A highlight optimization module is used to identify the highlight areas of the first face image and optimize the highlight areas to obtain a second face image.

[0049] The 3D mapping module is used to map the second face image onto the 3D head model of the target person, generating a texture map corresponding to the target person.

[0050] Compared with existing technologies, this application has the following beneficial effects: This application performs illumination normalization processing on the initial face image by using illumination information under pre-set standardized uniform illumination conditions to simulate an idealized illumination scene, thereby eliminating the difference in shadow depth caused by illumination condition deviations such as light source position shifts or intensity fluctuations, and achieving uniform shadow distribution; and by identifying and optimizing the highlight areas, it suppresses abnormal highlights caused by oil gloss reflection to obtain a second face image, and generates a texture map based on the second face image, thus achieving accurate optimization of shadow and highlight areas in the two-dimensional face image, avoiding the influence of irregular shadow and highlight areas on the rendering effect, thereby improving the rendering effect of the corresponding three-dimensional face model. Attached Figure Description

[0051] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0052] Figure 1 A schematic flowchart illustrating a method for generating a texture map according to an embodiment of this application;

[0053] Figure 2 Example diagrams of the initial face image and the first face image provided in the embodiments of this application;

[0054] Figure 3 A schematic diagram illustrating the process of obtaining the highlighted area of ​​the first face image provided in an embodiment of this application;

[0055] Figure 4 A schematic flowchart illustrating another method for generating texture maps provided in this application embodiment;

[0056] Figure 5 Example diagrams of a segmented mask image of a face region, a face region, and a repaired first face image provided for embodiments of this application;

[0057] Figure 6 Example diagram of the highlighted area and the second face image provided in the embodiments of this application;

[0058] Figure 7 A comparison example of texture maps and 3D face models provided in embodiments of this application;

[0059] Figure 8 This is a schematic diagram of a texture mapping generation device provided in an embodiment of this application. Detailed Implementation

[0060] As described earlier, under natural lighting conditions, 2D facial images often produce irregular local shadows due to differences in light source direction and intensity, or environmental reflections. Furthermore, local highlights (e.g., reflections on the forehead and tip of the nose) may occur due to sebum secretion on the facial skin or direct sunlight. Current techniques typically directly map the pixel colors of the original 2D facial image onto the surface of a 3D facial model. This results in irregular shadows and highlights remaining in the texture map, leading to poor rendering of the final 3D facial model. Furthermore, it can even destroy the three-dimensionality of the 3D facial model and produce inconsistent rendering effects under different lighting conditions.

[0061] For example, if the light source is too concentrated or too intense, it will create excessively deep shadows on the face, which will obscure facial details. Conversely, if the light source is too diffuse or too weak, the shadows will be too faint, failing to reflect the three-dimensional structure of the face. Therefore, complex lighting layouts or environmental reflections can cause uneven shadow distribution, resulting in some areas of the face having too faint shadows while others have too dark shadows. This uneven shadow distribution disrupts the overall harmony of the 3D face model. Furthermore, highlighted areas often have abnormally high brightness values, causing the color information in these areas to deviate significantly from realistic skin tones, and even resulting in unnatural color spots. This prevents texture maps from accurately reflecting the true colors of the face, leading to poor rendering and low realism of the 3D face model.

[0062] Furthermore, traditional AR / VR digital human rendering typically uses preset, fixed lighting parameters, which cannot adapt to the complex lighting conditions of real-world photography environments (such as natural light, indoor lighting, and reflected light). This leads to a mismatch between the rendering and the real-world lighting, potentially resulting in unnatural phenomena such as "floating" or "distortion." To address the effects of abnormal shadows and highlights during rendering, it is necessary to select an appropriate network structure and use deep learning-based lighting estimation algorithms to predict lighting parameters and dynamically match the real-world lighting. These lighting parameters also need to be integrated into the rendering pipeline. However, this approach faces significant challenges, such as requiring a lightweight network structure and seamless integration of lighting parameters into the rendering pipeline. Therefore, it is difficult to eliminate the effects of shadows and highlights during AR / VR digital human rendering.

[0063] This application provides a method for generating texture maps, including: acquiring an initial face image of a target person; performing illumination normalization processing on the initial face image based on preset illumination information to obtain a first face image; identifying the highlight areas of the first face image and optimizing the highlight areas to obtain a second face image; mapping the second face image onto a 3D head model of the target person to generate a texture map corresponding to the target person. In this application, the initial face image is illuminated and normalized using pre-set illumination information under standardized uniform illumination conditions to simulate an idealized illumination scene, thereby eliminating shadow depth differences caused by illumination condition deviations such as light source position shifts or intensity fluctuations, and achieving uniform shadow distribution; furthermore, by identifying and optimizing the highlight areas, abnormal highlights caused by gloss reflection are suppressed to obtain the second face image, and a texture map is generated based on the second face image. This application achieves precise optimization of shadow and highlight areas in a 2D face image, avoiding the influence of irregular shadow and highlight areas on the rendering effect, thereby improving the rendering effect of the corresponding 3D face model.

[0064] Furthermore, in the texture mapping generation method provided in this application, the optimization of shadow and highlight areas in the two-dimensional face image is carried out between texture mapping generation, that is, before rendering the three-dimensional face model. The influence of irregular shadows and highlights is eliminated through the preprocessing of the two-dimensional face image, thereby improving the rendering effect of the three-dimensional face model.

[0065] To enable those skilled in the art to better understand the present application, the technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present application, and not all embodiments. Based on the embodiments in the present application, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the scope of protection of the present application.

[0066] Example 1:

[0067] The following is combined with Figures 1-3 This paper will provide a detailed description of a texture mapping generation method provided in the embodiments of this application.

[0068] like Figure 1 As shown in the figure, a method for generating a texture map provided in this application includes the following steps:

[0069] S101. Obtain the initial facial image of the target person.

[0070] The initial face image is the raw, unprocessed 2D face image. For example, it is a 2D face image containing facial information of the target person, captured under normal lighting conditions (such as natural light, indoor lighting, etc.). The initial face image may include: facial shadows (such as the sides of the nose, eye sockets, etc.) caused by uneven light source direction, intensity, or environmental reflection, and facial highlights (such as the forehead, tip of the nose, chin, etc.) caused by sebum secretion, direct sunlight, or material reflection.

[0071] Furthermore, the initial face image contains the complete face region (usually a frontal or near-frontal view).

[0072] In one possible implementation, the foreground of the initial face image contains the face, while the background of the initial face image is black. The foreground refers to the main target or object of interest in the image (i.e., the subject), which is the part that needs to be processed, analyzed, or extracted; the background refers to the area in the image other than the foreground, which usually exists as a backdrop and environment.

[0073] S102. Based on the preset illumination information, the initial face image is subjected to illumination normalization processing to obtain the first face image.

[0074] The preset lighting information refers to the lighting information corresponding to standardized uniform lighting conditions set in advance. In other words, the preset lighting information refers to a set of predefined lighting parameters (including: light source direction, intensity, color temperature, etc.), which correspond to an idealized uniform lighting scene.

[0075] For example, the preset lighting information may include: the light source direction is 45° directly in front, simulating natural light to evenly cover the face; the light intensity is medium brightness to avoid overexposure or underexposure; and the color temperature is 5600K (sunlight color temperature) to reduce color cast.

[0076] The first face image is an image containing the face of the target person under uniform lighting conditions.

[0077] Specifically, based on preset illumination information, the algorithm adjusts the illumination conditions in the initial face image to the illumination conditions corresponding to the preset illumination information (adjusted to the ideal state), that is, adjusts the illumination conditions of the initial face image to uniform illumination conditions, so as to eliminate the problem of uneven shadows caused by differences in shooting environment (such as direct strong light source, indoor light reflection, etc.), and obtains the first face image.

[0078] In natural shooting environments, factors such as the position, intensity, and reflection path of light sources can lead to uneven shadow distribution on the face (e.g., strong light from one side creates a deep shadow, while multiple light sources create complex superpositions). For example, areas such as the forehead, bridge of the nose, and chin may exhibit discontinuous shadows with alternating light and dark areas due to differences in the angle of the light source. By normalizing the initial facial image using pre-set, standardized, and uniform lighting conditions, an idealized lighting scenario can be simulated. In this idealized lighting scenario, the angle and intensity of the light source illuminating the face remain consistent, avoiding differences in shadow depth caused by deviations in lighting conditions such as light source position shifts or intensity fluctuations, ultimately achieving uniform shadow distribution.

[0079] In one possible implementation, based on preset lighting information, the lighting conditions of the initial face image are adjusted to the lighting conditions corresponding to the preset lighting information using a diffusion-based lighting editing algorithm (IC-Light algorithm) to obtain the first face image.

[0080] IC-Light (Imposing Consistent Light) is a diffusion-based lighting editing algorithm that aims to capture the lighting information of the background image and adjust the lighting conditions of the subject (i.e., the foreground) to match the background lighting. In this embodiment, the IC-Light algorithm is used to adjust the lighting conditions of the subject based on the lighting information. This allows the subject (i.e., the face) to be adjusted to the preset lighting conditions while preserving its features, eliminating the interference of shadows and highlights, thereby improving the quality of the face image.

[0081] To make it easier to understand, the following will be combined with... Figure 2 This example illustrates the comparison between the initial face image and the first face image. For clarity, a portion of the face image will be used as an example for a simplified description.

[0082] like Figure 2 In the initial face image shown in (a) (which can be referred to as the partial initial face image), the lighting conditions are those of a natural shooting environment. There is a significant problem of uneven shadow distribution on the face surface, such as darker shadows in the forehead area, which obscure facial details. Furthermore, the face is generally dark, lacks brightness, and has a flat visual effect.

[0083] like Figure 2 In a portion of the first face image shown in (b) (which can be referred to as the partial first face image), the lighting conditions are those corresponding to the preset lighting information (i.e., uniform lighting conditions). The shadows on the face are evenly and softly distributed, with no obviously too dark or too light areas, reducing lighting interference. Furthermore, the brightness of the face is significantly improved, the overall face is more prominent, and facial details are clearer.

[0084] S103. Identify the highlighted areas of the first face image and optimize the highlighted areas to obtain the second face image.

[0085] Obtaining the first face image involves adjusting the overall lighting conditions to match the preset lighting information. The optimization focuses on eliminating shadows and improving overall brightness. Abnormal highlights caused by oil reflection are usually caused by the oil layer on the skin surface or direct exposure to strong light sources, manifesting as abnormally high brightness in local pixels (such as reflections on the forehead, nose tip, and chin). These bright areas are non-uniform, with small and concentrated areas of oil reflection that are not directly related to the overall lighting conditions. Furthermore, oil reflection is a direct result of surface reflection and is unrelated to skin texture, requiring local processing to resolve.

[0086] In summary, adjusting global illumination conditions makes it difficult to accurately locate and eliminate local highlights; that is, abnormal highlights caused by oil reflection in two-dimensional face images are not effectively suppressed. Therefore, in this embodiment, the highlight areas of the first face image are identified and optimized to obtain the second face image, thereby suppressing abnormal highlights caused by oil reflection.

[0087] In one possible implementation, a first highlight region of the first face image is identified based on the saturation distribution corresponding to the first face image; a second highlight region of the first face image is identified based on the RGB channel difference corresponding to the first face image; and a highlight region of the first face image is generated based on the first and second highlight regions.

[0088] To make it easier to understand, the following will be combined with... Figure 3 This section details how to obtain the highlighted areas of a first-person face image.

[0089] S301. Based on the first face image, generate the corresponding saturation channel map.

[0090] Specifically, the first image is converted from the RGB color space to the HSV color space, and the saturation channel map is extracted.

[0091] The RGB (Red, Green, Blue) color space is a color model based on additive color mixing, which represents colors through linear combinations of the three primary colors: red (R), green (G), and blue (B).

[0092] The HSV (Hue, Saturation, Value) color space is a color model based on hue (H), saturation (S), and value (V). Hue represents the color type, saturation represents the color purity, and value represents the color lightness.

[0093] In this context, the highlight area can be understood as the white area. Therefore, the lower the saturation of a pixel, the closer it is to white, and the more it belongs to the highlight area; conversely, the higher the saturation of a pixel, the further it is from white, and the less it belongs to the highlight area.

[0094] S302. Perform histogram analysis on the saturation channel map to determine the saturation threshold.

[0095] Histogram analysis is a method of studying the distribution characteristics of data by drawing and interpreting histograms. By grouping data according to certain rules and counting the frequency or density of data in each group, it can intuitively show the central tendency, dispersion and distribution pattern of the data.

[0096] Specifically, a histogram analysis is performed on the saturation channel map to statistically analyze the distribution of pixel saturation values, determine the peak range of low-saturation regions, and select a lower saturation value as the saturation threshold. For example, a histogram analysis of the saturation channel map shows that the saturation of most pixels is concentrated between 30 and 255, while the low-saturation region (S<30) accounts for a small proportion; therefore, the threshold can be set to 30.

[0097] Since the saturation distribution varies greatly among different images, a fixed saturation threshold may lead to over-segmentation or under-segmentation. Therefore, the saturation threshold is dynamically determined (i.e., the saturation threshold is determined using histogram analysis).

[0098] S303. Based on the saturation threshold, select pixels in the saturation channel image whose saturation value is less than the saturation threshold to obtain the first filtering area.

[0099] Specifically, when the saturation value of a pixel is less than the saturation threshold, the pixel is considered to belong to the highlight area. Then, pixels with saturation values ​​lower than the saturation threshold in the saturation channel image are selected to obtain the first filter area.

[0100] S304. Perform erosion and dilation operations on the first selected area to obtain the first highlighted area of ​​the first face image.

[0101] Erosion is an operation that shrinks the foreground region by removing pixels on the boundary, causing the target region to shrink inward. A structuring element (such as a 3x3 matrix, a cross, etc.) is slid across the image; if the structuring element completely covers the pixels of the target region, the center point is preserved; otherwise, the center point is set as the background.

[0102] Specifically, an erosion operation is performed on the first selected area to remove small-area noise (such as misidentified pixels), isolated points, or minor noise within the first selected area. For example, in the first selected area, low-saturation noise may be incorrectly identified as bright areas; the erosion operation can remove it.

[0103] Dilation is an operation that expands the foreground region by adding pixels to the boundary, causing the target region to expand outward. A structuring element is slid across the image; if the resulting element overlaps with the target region, its center point is set to the foreground; otherwise, the original value is retained.

[0104] Specifically, by expanding the first filtering region, adjacent first filtering regions are merged into a single complete region, reducing false positives. For example, in the first filtering region obtained, the highlighted areas formed by reflections from the tip of the nose and forehead may be divided into multiple small regions; the expansion operation can merge them into a continuous region.

[0105] By performing erosion and dilation operations on the first screening area, noise in the first screening area is removed by erosion, and broken areas in the first screening area are connected by dilation, thereby obtaining a more accurate first highlight area.

[0106] The above S301-S304 details how to obtain the first bright area. The following, combined with S305-S306, details how to obtain the second bright area.

[0107] S305. Based on the RGB channel difference corresponding to the first face image, select pixels whose RGB channel difference is less than the preset RGB channel difference threshold to obtain the second filtering area.

[0108] In particular, the highlighted area can be understood as the white area. The RGB channel data are similar, that is, the smaller the difference between the RGB channels, the more likely it is to be white.

[0109] Specifically, for each pixel in the first face image, the difference between its red, green, and blue channels is calculated. For example, the first difference is the difference between the red and green channels, the second difference is the difference between the red and blue channels, and the third difference is the difference between the green and blue channels. Based on the differences between the red, green, and blue channels, the RGB channel difference for each color is determined. This can be either the maximum value among the three differences or the sum of the three differences.

[0110] S306. Perform an erosion and expansion operation on the second screening area to obtain the second highlighted area.

[0111] By performing erosion and dilation operations on the second filtering area, noise in the second filtering area is removed by erosion, and broken areas in the second filtering area are connected by dilation, thereby obtaining a more accurate second highlight area. This avoids inaccurate highlight area determination due to the influence of preset RGB channel difference threshold, noise, etc.

[0112] S307. Perform erosion and expansion operations on the first and second highlighted areas to obtain the highlighted areas of the first face area.

[0113] By performing further erosion and dilation operations on the first and second selection regions, noise is removed from both regions through erosion, and broken areas are connected through dilation, resulting in accurate highlighted areas. Furthermore, dilation of the first and second selection regions covers a larger area as the highlighted area, ensuring that all highlighted areas in the first face region are covered. This prevents subsequent omissions of highlighted areas that would not be optimized, thus avoiding abnormally bright areas in the final second face region.

[0114] It should be noted that S301-S304 and S305-S306 can be performed sequentially or simultaneously. That is, the first highlighted area can be obtained first and then the second highlighted area can be obtained; the second highlighted area can be obtained first and then the first highlighted area can be obtained; or the first highlighted area and the second highlighted area can be obtained simultaneously or almost simultaneously. No specific limitation is made in this application.

[0115] The above combination Figure 3 This section details how to obtain the highlighted areas of the first face image. The following section will continue with... Figure 1 This application introduces a method for generating texture maps according to an embodiment.

[0116] In one possible implementation, the highlighted areas are optimized using a lighting optimization algorithm to obtain the second face image.

[0117] The goal of the illumination optimization algorithm is to reduce the brightness of bright areas while maintaining the natural flow of the image and avoiding artifacts or distortion.

[0118] Specifically, the brightness and saturation of the highlighted areas are reduced through a lighting optimization algorithm; and feathering technology is used at the boundary between the highlighted and non-highlighted areas to ensure the smoothness of the lighting adjustment.

[0119] For example, the lighting optimization algorithm could be the illumination change algorithm, which adjusts local contrast and maps lighting to reduce the brightness and saturation of bright areas while preserving skin texture details.

[0120] Feathering is an edge transition technique in image processing that creates a transition zone (such as a feathery edge) with decreasing force at the edges of a selected area, allowing the selected area to blend naturally with the background or other areas. For example, Gaussian blur is a common implementation of feathering, using a Gaussian blur algorithm to apply a weighted average to the edges of the selected area, causing the edge pixels to gradually transition into the background.

[0121] Furthermore, feathering technology can be combined with progressive lighting mapping to further optimize the transition effect by adjusting the lighting intensity or color distribution in the edge areas.

[0122] Specifically, the lighting optimization algorithm is used to optimize the bright areas to avoid skin tone distortion (such as graying) caused by excessive suppression of brightness and saturation; feathering technology is used to avoid hard edges that may appear at the boundary between bright and non-bright areas.

[0123] S104. Map the second face image onto the 3D head model of the target person to generate a texture map corresponding to the target person.

[0124] Among them, the 3D head model is a 3D head model of the target person obtained by 3D reconstruction based on the initial face image.

[0125] Specifically, based on a differentiable rendering tool library (e.g., nvdiffrast), the second face image is mapped onto the 3D head model of the target person to generate a texture map corresponding to the target person.

[0126] This application provides a method for generating texture maps, comprising: acquiring an initial face image of a target person; performing illumination normalization processing on the initial face image based on preset illumination information to obtain a first face image; identifying the highlight areas of the first face image and optimizing the highlight areas to obtain a second face image; and mapping the second face image onto a 3D head model of the target person to generate a texture map corresponding to the target person. In this application embodiment, the initial face image is normalized using illumination information under preset standardized uniform illumination conditions to simulate an idealized illumination scene, thereby eliminating shadow depth differences caused by illumination condition deviations such as light source position shifts or intensity fluctuations, achieving uniform shadow distribution; and by identifying and optimizing the highlight areas, abnormal brightness caused by gloss reflection is suppressed to obtain the second face image, and a texture map is generated based on the second face image. This achieves precise optimization of shadow and highlight areas in the 2D face image, avoiding the influence of irregular shadow and highlight areas on the rendering effect, thereby improving the rendering effect of the corresponding 3D face model.

[0127] Furthermore, in the texture mapping generation method provided in this application embodiment, the optimization of shadow and highlight areas in the two-dimensional face image is carried out between texture mapping generation, that is, before rendering the three-dimensional face model. The influence of irregular shadows and highlights is eliminated through preprocessing of the two-dimensional face image, thereby improving the rendering effect of the three-dimensional face model.

[0128] Example 2:

[0129] The following is combined with Figures 4-7 This paper details another method for generating texture maps provided in the embodiments of this application.

[0130] S401. Obtain the initial facial image of the target person.

[0131] S402. Based on the preset illumination information, the initial face image is subjected to illumination normalization processing to obtain the first face image.

[0132] It should be noted that S401 and S402 are the same as S101 and S102 in Embodiment 1. Therefore, for details of the specific implementation of S401 and S402, please refer to S101 and S102 in Embodiment 1, and will not be repeated here.

[0133] S403. Identify the face region and background region in the first face image.

[0134] Specifically, based on the first face image, a face region segmentation mask is generated using a face segmentation algorithm; based on the face region segmentation mask, the face region in the first face image is identified; the face region segmentation mask is inversely selected to generate a background region segmentation mask; based on the background region segmentation mask, the background region in the first face image is identified.

[0135] Face segmentation is used to accurately separate facial regions (including skin, facial features, etc.) from the background (or other objects) in an image and output a binary mask image.

[0136] Specifically, a face segmentation algorithm generates a binary mask (a segmentation mask of the face region) containing the main facial features (skin, eyebrows, eyes, etc.) and the background, where 1 represents the face region and 0 represents the background region. This face region segmentation mask is then applied to a first face image to identify the face region. The face region segmentation mask is then inverted to generate a background region segmentation mask, where 1 represents the background region and 0 represents the face region. This background region segmentation mask is then applied to a second face image to identify the background region.

[0137] S404. Based on the face region, perform image restoration on the background region to obtain the restored background region.

[0138] Inaccurate facial region recognition can lead to blurred or broken edges, especially around hair, beards, skin, and the boundary between the face and the background. For example, in the first face image, the face edge may be similar in color to the background (e.g., gray hair and a gray wall), causing the recognized face region to actually include the background of a similar color. This affects the subsequent texture mapping (e.g., the texture map includes elements from the background region), thus impacting the rendering of the 3D face model.

[0139] Therefore, based on the face region, image inpainting is performed on the background region to make the boundary pixels of the repaired background region and the face region transition naturally and eliminate the influence of background pixels on the face edge.

[0140] Image inpainting is a technique that uses algorithms to reconstruct lost, damaged, or missing parts of an image, aiming to make the restored image visually coherent and natural.

[0141] For example, the background near the face boundary is replaced with reasonable content to avoid residual background pixels in the texture; and by expanding the boundary of the segmentation mask map of the face region, the wider background area is further patched to ensure that the face edge pixels are completely covered.

[0142] S405. Based on the repaired background area and face area, generate the repaired first face image.

[0143] Specifically, the repaired background area and face area are combined to generate the repaired first face image.

[0144] To make it easier to understand, the following will be combined with... Figure 5 Examples are given to illustrate the segmentation mask of the face region, the face region, and the restored first face image.

[0145] The segmentation mask of the face region in the first face image is as follows: Figure 5 As shown in (a) in the figure, for ease of explanation, it is as follows Figure 5 Taking a portion of the face region segmentation mask shown in (a) as an example (which can be simply referred to as the partial face region segmentation mask), a brief description will be given. The partial face region segmentation mask is as follows: Figure 5 As shown in (b) of the diagram.

[0146] Will as Figure 5 The partial segmentation mask map of the face region shown in (b) is applied to the first face image to identify a portion of the face region (which can be simply referred to as the partial face region). Figure 5 As shown in (c) above. A segmentation mask for the background region of the first face image is generated by inverting the segmentation mask of the face region, and the background region of the first face image is identified based on the segmentation mask of the background region; image inpainting is performed on the background region based on the face region to obtain the repaired background region; a repaired first face image is generated based on the face region and the repaired background region, wherein a portion of the repaired first face image (which can be simply referred to as the repaired partial first face image) is shown below. Figure 5 As shown in (d) in the figure.

[0147] like Figure 5 The image (a) shows a segmentation mask of the face region, where the white part represents the face region and the black part represents the background region.

[0148] like Figure 5 The partial segmentation mask of the face region shown in (b) is shown in the figure, where the white part represents the face region and the black part represents the background region.

[0149] like Figure 5 The partial face region shown in (c) of the first face image only retains the facial features. Figure 5 The face region corresponding to the white mask in (b) of the image.

[0150] like Figure 5 The restored portion of the first face image shown in (d) shows the background area, which is naturally integrated with the face area after image restoration processing.

[0151] S406. Identify the highlighted areas of the repaired first face image and optimize the highlighted areas to obtain the second face image.

[0152] To make it easier to understand, the following will be combined with... Figure 6 Examples are given for the highlighted areas of the restored first face image and the second face image. For ease of explanation, the following description uses a portion of the highlighted area of ​​the restored first face image (which can be simply referred to as a portion of the highlighted area of ​​the restored first face image) and a portion of the second face image (which can be simply referred to as a portion of the second face image) as examples. The restored portion of the first face image is shown below. Figure 5 As shown in (d) in the figure.

[0153] like Figure 6 The partial highlight area of ​​the restored first face image shown in (a) is an example, where the white areas are highlighted areas and the black areas are non-highlighted areas. The image shows a portion of the restored first face image with highlights such as... Figure 6 The highlighted area shown in (a) is optimized to obtain a partial second face image as shown below. Figure 6 As shown in (b) above, compared to Figure 5 In (d), the bright areas caused by oil reflection are effectively suppressed and smoothly transitioned with the surrounding areas.

[0154] It should be noted that S406 is the same as S103 in Implementation 1. Therefore, for details on the specific implementation of S406, please refer to S103 in Implementation 1, which will not be repeated here.

[0155] S407. Perform facial landmark detection on the second face image to obtain the coordinates of the facial landmarks.

[0156] Among them, facial landmark detection is a computer vision technology that uses algorithms to identify specific locations of a face in an image (including eyes, nose tip, corners of mouth, etc.) and outputs a set of coordinates for those locations to locate key parts of the face.

[0157] The facial key point coordinates include: corner of the mouth key point coordinates, cheek key point coordinates, and eye key point coordinates. In the embodiments of this application, the main facial key point coordinates used are the corner of the mouth key point coordinates and the cheek key point coordinates.

[0158] The key coordinates of the corners of the mouth include: the key coordinates of the left inner corner of the mouth, the key coordinates of the left outer corner of the mouth, the key coordinates of the right inner corner of the mouth, and the key coordinates of the right outer corner of the mouth.

[0159] S408. Based on the coordinates of key points at the corners of the mouth, determine the shadow area at the corners of the mouth in the second face image.

[0160] Specifically, the key point coordinates for the corners of the mouth include: the left inner corner of the mouth, the left outer corner of the mouth, the right inner corner of the mouth, and the right outer corner of the mouth. The corresponding shading areas for the corners of the mouth include: the left corner shading area and the right corner shading area.

[0161] For the left corner of the mouth, the shaded area is defined as follows: the distance between the coordinates of the left inner corner of the mouth and the key point of the left outer corner of the mouth is used as the radius, and the coordinates of the left outer corner of the mouth is used as the center. Similarly, the shaded area for the right corner of the mouth is defined as follows: the distance between the coordinates of the right inner corner of the mouth and the key point of the right outer corner of the mouth is used as the radius, and the coordinates of the right outer corner of the mouth is used as the center.

[0162] S409. Based on the coordinates of cheek key points, determine the cheek replacement region in the second face image that has the same area as the shadow area at the corner of the mouth.

[0163] Specifically, the cheek region is determined based on the coordinates of key cheek points, and an area with the same size as the corner shadow area is selected within the cheek region as the cheek replacement area.

[0164] S410. Based on the cheek replacement area, perform pixel replacement processing on the shadow area at the corner of the mouth to obtain the replaced second face image.

[0165] Due to the structure of the human face, there are usually noticeable indentations or contours around the corners of the mouth, which can cause shadows to form under certain lighting conditions. Furthermore, when making certain facial expressions (such as smiling), the corners of the mouth stretch, potentially exacerbating the shadow effect in this area. Therefore, based on the cheek replacement area, pixel replacement processing is performed on the shadow area around the corners of the mouth to remove or reduce the visual interference caused by the shadows, making the skin appear smoother and more even, and improving overall aesthetics.

[0166] In one possible implementation, the pixel values ​​of the cheek replacement region are mapped to the corner shadow region of the mouth; the mapped corner shadow region of the mouth is smoothed using a Poisson fusion algorithm to obtain the replaced second face image.

[0167] Poisson fusion, in particular, achieves seamless image fusion using the Poisson equation in mathematics. It preserves the gradient information (such as edge and texture details) of the original image while matching the boundary conditions (such as lighting and color) of the target image. Specifically, Poisson fusion allows the replacement cheek area to transition naturally with the surrounding image of the shadow area at the corner of the mouth, avoiding obvious boundary marks and resulting in a second face image with the replacement, thus eliminating the shadow area at the corner of the mouth.

[0168] S411. Map the replaced second face image onto the 3D head model of the target person to generate a texture map corresponding to the target person.

[0169] To make it easier to understand, the following will be combined with... Figure 7 Let's take an example to illustrate the texture map corresponding to the target character. For ease of explanation, we will use a portion of the texture map (which can be simply referred to as a partial texture map) as an example below for a brief explanation.

[0170] like Figure 7 The partial texture map obtained in the embodiment of this application shown in (a) has significantly improved the absence of abnormal shadows and highlights on the human face, and the transition of light and shadow in each area of ​​the face is natural, avoiding local over-brightness or under-brightness; the details such as skin texture and pores are clearer, enhancing the realism; and the overall skin tone is uniform, without obvious color block segmentation, improving visual harmony.

[0171] And such Figure 7 In the prior art shown in (b), some texture maps have abnormal shadows and highlights on the human face, and there is obvious uneven lighting. Some areas appear too bright or too dark, which affects the overall appearance. In addition, there are color differences between different areas, resulting in inconsistent skin tones.

[0172] Similarly, Figure 7 A portion of the 3D face model shown in (c) (which can be simply referred to as the partial 3D face model) is based on, for example... Figure 7 The partial texture map rendering obtained in the embodiment of this application shown in (a) has natural changes in facial lighting under illumination conditions. The highlights are soft and accurately positioned, while the low-light areas are rich in detail, enhancing the sense of three-dimensionality. Furthermore, the face blends well with the background, clothing, and other elements without any abruptness, thus improving the overall rendering effect of the three-dimensional face model.

[0173] And such Figure 7 The partial 3D face model shown in (d) is based on, for example, Figure 7The partial texture mapping obtained in the prior art shown in (b) results in a harsh lighting effect, with highlights that may be too glaring and shadows that appear dull. There are abnormal highlights and shadows, color differences between different areas, leading to inconsistent skin tones and obvious color spots. Furthermore, the integration between the face and elements such as the background and clothing is poor, making it appear unnatural and resulting in poor rendering of the 3D face model.

[0174] This application provides a method for generating texture maps, comprising: acquiring an initial face image of a target person; performing illumination normalization processing on the initial face image based on preset illumination information to obtain a first face image; identifying the face region and background region in the first face image; performing image restoration on the background region based on the face region to obtain a restored background region; generating a restored first face image based on the restored background region and face region; identifying the highlight region of the restored first face image and optimizing the highlight region to obtain a second face image; performing facial key point detection on the second face image to obtain the coordinates of the facial key points; determining the corner-of-mouth shadow region in the second face image based on the coordinates of the corner-of-mouth key points; determining a cheek replacement region in the second face image with the same area as the corner-of-mouth shadow region based on the coordinates of the cheek key points; performing pixel replacement processing on the corner-of-mouth shadow region based on the cheek replacement region to obtain a replaced second face image; and mapping the replaced second face image onto a 3D head model of the target person to generate a texture map corresponding to the target person. In this embodiment, not only is precise optimization of shadow and highlight areas in a 2D face image achieved, avoiding the impact of irregular shadow and highlight areas on the rendering effect; the background area of ​​the first face image is also repaired, making the boundary pixel transition between the repaired background area and the face area natural, thereby eliminating the influence of background pixels on the face edge; furthermore, based on the cheek replacement area, pixel replacement processing is performed on the corner shadow area of ​​the mouth to remove or reduce the visual interference caused by the corner shadow, making the skin look smoother and more uniform, improving the overall aesthetics, and improving the rendering effect of the texture map, thereby improving the rendering effect of the corresponding 3D face model.

[0175] Example 3:

[0176] The following is combined with Figure 8 This application provides a detailed description of a texture mapping generation apparatus according to its embodiments.

[0177] like Figure 8 As shown in the embodiment of this application, a texture mapping generation apparatus includes the following modules:

[0178] Image acquisition module 801 is used to acquire the initial face image of the target person;

[0179] The illumination processing module 802 is used to perform illumination normalization processing on the initial face image based on preset illumination information to obtain the first face image; wherein, the preset illumination information is the illumination information corresponding to the pre-set standardized uniform illumination conditions.

[0180] The highlight optimization module 803 is used to identify the highlight areas of the first face image and optimize the highlight areas to obtain the second face image;

[0181] The 3D mapping module 804 is used to map the second face image onto the 3D head model of the target person to generate a texture map corresponding to the target person.

[0182] In one possible implementation, the highlight optimization module 803 includes: a first highlight recognition module, a second highlight recognition module, and a highlight area generation module.

[0183] The first highlight recognition module is used to identify the first highlight area of ​​the first face image based on the saturation distribution corresponding to the first face image;

[0184] The second highlight recognition module is used to identify the second highlight area of ​​the first face image based on the RGB channel difference corresponding to the first face image;

[0185] The highlight region generation module is used to generate the highlight region of the first face image based on the first highlight region and the second highlight region.

[0186] In one possible implementation, the first highlight recognition module is specifically used to generate a corresponding saturation channel map based on the first face image; perform histogram analysis on the saturation channel map to determine the saturation threshold; based on the saturation threshold, filter out pixels in the saturation channel map whose saturation value is less than the saturation threshold to obtain a first filtering region; and perform erosion and dilation operations on the first filtering region to obtain the first highlight region of the first face image.

[0187] In one possible implementation, the second highlight recognition module is specifically used to filter out pixels whose RGB channel difference is less than a preset RGB channel difference threshold based on the RGB channel difference corresponding to the first face image, to obtain a second filtering region; and to perform erosion and dilation operations on the second filtering region to obtain a second highlight region of the first face image.

[0188] In one possible implementation, the highlight region generation module is specifically used to perform erosion and dilation operations on the first highlight region and the second highlight region to obtain the highlight region of the first face image.

[0189] In one possible implementation, the device further includes: a region recognition module, a background restoration module, and an image synthesis module.

[0190] The region recognition module is used to identify the face region and background region in the first face image;

[0191] The background restoration module is used to restore the background region based on the face region, and obtain the restored background region.

[0192] The image synthesis module is used to generate a restored first face image based on the restored background area and face area.

[0193] The highlight optimization module 803 is specifically used to identify the highlight areas of the repaired first face image and optimize the highlight areas to obtain the second face image.

[0194] In one possible implementation, the region recognition module is specifically used to generate a segmentation mask map of the face region based on the first face image using a face segmentation algorithm; identify the face region in the first face image based on the segmentation mask map of the face region; inversely select the segmentation mask map of the face region to generate a segmentation mask map of the background region; and identify the background region in the first face image based on the segmentation mask map of the background region.

[0195] In one possible implementation, the device further includes: a key point detection module, a corner-of-mouth shadow recognition module, a replacement area recognition module, and a shadow replacement module.

[0196] The key point detection module is used to detect facial key points in the second face image and obtain the coordinates of the facial key points; among which, the coordinates of the facial key points include: the coordinates of the corner of the mouth key points and the coordinates of the cheek key points;

[0197] The corner-of-mouth shadow recognition module is used to determine the corner-of-mouth shadow area in the second face image based on the coordinates of key points at the corner of the mouth.

[0198] The replacement region recognition module is used to determine the cheek replacement region with the same area as the corner shadow region in the second face image based on the coordinates of cheek key points;

[0199] The shadow replacement module is used to perform pixel replacement on the shadow area at the corner of the mouth based on the cheek replacement area, so as to obtain a second face image after replacement.

[0200] In one possible implementation, the shadow replacement module is specifically used to map the pixel values ​​of the cheek replacement area to the corner shadow area of ​​the mouth; the Poisson fusion algorithm is used to smooth the mapped corner shadow area of ​​the mouth to obtain the replaced second face image.

[0201] This application provides a texture mapping generation device, comprising: an image acquisition module 801 for acquiring an initial face image of a target person; a lighting processing module 802 for performing lighting normalization processing on the initial face image based on preset lighting information to obtain a first face image; a highlight optimization module 803 for identifying the highlight areas of the first face image and optimizing the highlight areas to obtain a second face image; and a three-dimensional mapping module 804 for mapping the second face image onto a three-dimensional head model of the target person to generate a texture mapping corresponding to the target person. In this embodiment, the initial face image is normalized using pre-set standardized uniform lighting information to simulate an idealized lighting scene. This eliminates shadow depth differences caused by lighting condition deviations such as light source position shifts or intensity fluctuations, achieving uniform shadow distribution. Furthermore, by identifying and optimizing highlight areas, abnormal highlights caused by oily reflections are suppressed, resulting in a second face image. A texture map is then generated based on the second face image. This achieves precise optimization of shadow and highlight areas in the two-dimensional face image, avoiding the impact of irregular shadow and highlight areas on the rendering effect, thereby improving the rendering effect of the corresponding three-dimensional face model.

[0202] Furthermore, in this embodiment, the optimization of shadow and highlight areas in the two-dimensional face image is performed between the generation of texture maps, that is, before rendering the three-dimensional face model. The preprocessing of the two-dimensional face image eliminates the influence of irregular shadows and highlights, thereby improving the rendering effect of the three-dimensional face model.

[0203] It should be noted that the various embodiments in this specification are described in a progressive manner, and the same or similar parts between the various embodiments can be referred to mutually. Each embodiment focuses on describing the differences from other embodiments. In particular, for the device embodiments, since they are basically similar to the method embodiments, the description is relatively simple, and the relevant parts can be referred to the description of the method embodiments. The device embodiments described above are merely illustrative, and the units described as separate components may or may not be physically separate. The components indicated as units may or may not be physical units, that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment solution according to actual needs. Those skilled in the art can understand and implement this without creative effort.

[0204] The above description is merely one specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the technical scope disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

Claims

1. A method for generating texture maps, characterized in that, include: Obtain the initial facial image of the target person; Based on preset illumination information, the initial face image is subjected to illumination normalization processing to obtain a first face image; wherein, the preset illumination information is the illumination information corresponding to a pre-set standardized uniform illumination condition; Identify the highlighted areas of the first face image and optimize the highlighted areas to obtain the second face image; The second face image is mapped onto the 3D head model of the target person to generate a texture map corresponding to the target person; The process of identifying the highlighted areas of the first face image and optimizing those areas to obtain the second face image includes: The highlighted areas of the repaired first face image are identified and optimized to obtain a second face image. The repaired first face image is obtained by repairing the background area based on the face area in the first face image.

2. The method according to claim 1, characterized in that, The identification of the highlighted areas of the first face image includes: Based on the saturation distribution corresponding to the first face image, identify the first bright area of ​​the first face image; Based on the RGB channel difference corresponding to the first face image, identify the second highlight area of ​​the first face image; Based on the first highlighted area and the second highlighted area, the highlighted area of ​​the first face image is generated.

3. The method according to claim 2, characterized in that, The step of identifying the first highlighted region of the first face image based on the saturation distribution corresponding to the first face image includes: Based on the first face image, generate the corresponding saturation channel map; Histogram analysis is performed on the saturation channel map to determine the saturation threshold; Based on the saturation threshold, pixels with saturation values ​​less than the saturation threshold in the saturation channel image are filtered out to obtain the first filtering region; An erosion and dilation operation is performed on the first selected area to obtain the first highlighted area of ​​the first face image.

4. The method according to claim 2, characterized in that, The step of identifying the second highlighted region of the first face image based on the RGB channel difference corresponding to the first face image includes: Based on the RGB channel difference corresponding to the first face image, pixels whose RGB channel difference is less than a preset RGB channel difference threshold are selected to obtain a second filtering area; An erosion and dilation operation is performed on the second selected area to obtain the second highlighted area of ​​the first face image.

5. The method according to claim 2, characterized in that, The step of generating the highlight region of the first face image based on the first highlight region and the second highlight region includes: Erosion and dilation operations are performed on the first and second highlighted areas to obtain the highlighted areas of the first face image.

6. The method according to claim 1, characterized in that, The restored first face image was obtained in the following way: Identify the face region and background region in the first face image; Based on the face region, the background region is inpainted to obtain the inpainted background region; Based on the repaired background area and the face area, a repaired first face image is generated.

7. The method according to claim 6, characterized in that, The identification of the face region and background region in the first face image includes: Based on the first face image, a segmentation mask map of the face region is generated using a face segmentation algorithm; Based on the segmentation mask map of the face region, identify the face region in the first face image; Inversely select the segmentation mask image of the face region to generate a segmentation mask image of the background region; Based on the segmentation mask map of the background region, the background region in the first face image is identified.

8. The method according to claim 1, characterized in that, The method further includes: Facial landmark detection is performed on the second face image to obtain the coordinates of the facial landmarks; wherein, the coordinates of the facial landmarks include: the coordinates of the corner of the mouth landmarks and the coordinates of the cheek landmarks; Based on the coordinates of the key points at the corners of the mouth, the shadow area at the corners of the mouth in the second face image is determined; Based on the coordinates of the cheek key points, determine the cheek replacement area in the second face image that has the same area as the shadow area at the corner of the mouth; Based on the cheek replacement area, pixel replacement processing is performed on the corner of the mouth shadow area to obtain the replaced second face image.

9. The method according to claim 8, characterized in that, The step of performing pixel replacement processing on the shadow area at the corner of the mouth based on the cheek replacement area to obtain the replaced second face image includes: Map the pixel values ​​of the cheek replacement area to the corner of the mouth shadow area; The Poisson fusion algorithm is used to smooth the shadow area at the corner of the mouth after mapping, so as to obtain the second face image after replacement.

10. A texture mapping generation apparatus, characterized in that, include: The image acquisition module is used to acquire the initial facial image of the target person; The illumination processing module is used to perform illumination normalization processing on the initial face image based on preset illumination information to obtain a first face image; wherein, the preset illumination information is the illumination information corresponding to a pre-set standardized uniform illumination condition; A highlight optimization module is used to identify the highlight areas of the first face image and optimize the highlight areas to obtain a second face image. The 3D mapping module is used to map the second face image onto the 3D head model of the target person, and generate a texture map corresponding to the target person; The highlight optimization module is specifically used to identify the highlight areas of the repaired first face image and optimize the highlight areas to obtain the second face image. The repaired first face image is obtained by repairing the background area based on the face area in the first face image.

Images