Texture generation method, system and storage medium for three-dimensional character model

Abstract

The present invention provides a texture generation method, system and storage medium for a three-dimensional character model, and relates to the field of image processing technology, including: performing multi-angle image acquisition on a target character to obtain a multi-angle character image set that meets a predetermined image quality constraint; performing visible texture feature detection according to a predetermined aiming frame and a visible texture detection channel to obtain a multi-angle visible texture feature distribution; performing three-dimensional modeling to generate a first texture model of the character; performing invisible texture feature compensation to generate a second texture model of the character; performing multi-angle UV expansion inspection to obtain a multi-angle UV expansion inspection result; and correcting and generating a third texture model of the character. The present invention solves the technical problem that in the generation of traditional three-dimensional character models, the limitation of the acquisition viewing angle causes a lack of texture information in some areas of the model surface, resulting in a lack of details in the texture of the generated model and low authenticity.

Description

Technical Field

[0001] The present invention relates to the technical field of image processing, and in particular to a texture generation method, system and storage medium for a three-dimensional character model. Background Art

[0002] The texture generation technology of 3D character models has been widely used in the fields of film and television production, game development, virtual reality, etc. These scenes have put forward higher and higher requirements on the authenticity and visual consistency of 3D character models, especially strict requirements on the details and integrity of textures. However, the existing technology usually relies on image acquisition from a single or a small number of angles, resulting in the lack of texture or insufficient information in invisible areas of the model (such as armpits, back, etc.). This defect directly leads to incomplete performance of the generated 3D texture model, which is difficult to meet the requirements of high-precision applications; in addition, the existing technology is prone to problems such as texture stretching, distortion or unnatural seams during UV unfolding and texture mapping, resulting in uneven texture on the surface of the model, especially on complex geometric surfaces, resulting in insufficient visual realism of the final model. Summary of the invention

[0003] The present application provides a texture generation method, system and storage medium for a three-dimensional character model, aiming to solve the technical problem in traditional three-dimensional character model generation that the limitation of the acquisition viewing angle causes the lack of texture information in some areas of the model surface, resulting in the lack of details and low authenticity of the texture of the generated model.

[0004] In a first aspect disclosed in the present application, a texture generation method for a three-dimensional character model is provided, the method comprising: performing multi-angle image acquisition on a target character to obtain a multi-angle character image set that satisfies a predetermined image quality constraint; performing visible texture feature detection on the multi-angle character image set according to a predetermined aiming frame and a visible texture detection channel to obtain a multi-angle visible texture feature distribution; performing three-dimensional modeling according to the multi-angle visible texture feature distribution to generate a first texture model of the character; performing invisible texture feature compensation on the first texture model of the character according to the multi-angle character image set to generate a second texture model of the character; performing a multi-angle UV unfolding test on the second texture model of the character according to the multi-angle character image set to obtain a multi-angle UV unfolding test result; and correcting the second texture model of the character according to the multi-angle UV unfolding test result to generate a third texture model of the character.

[0005] The second aspect disclosed in the present application provides a texture generation system for a three-dimensional character model, the system being used for the texture generation method for the above-mentioned three-dimensional character model, the system comprising: an image acquisition module, for performing multi-angle image acquisition of a target character to obtain a multi-angle character image set that satisfies a predetermined image quality constraint; a feature detection module, for performing visible texture feature detection on the multi-angle character image set according to a predetermined aiming frame and a visible texture detection channel to obtain a multi-angle visible texture feature distribution; a three-dimensional modeling module, for performing three-dimensional modeling according to the multi-angle visible texture feature distribution to generate a first texture model of the character; a feature compensation module, for performing invisible texture feature compensation on the first texture model of the character according to the multi-angle character image set to generate a second texture model of the character; an expansion inspection module, for performing a multi-angle UV expansion inspection on the second texture model of the character according to the multi-angle character image set to obtain a multi-angle UV expansion inspection result; and a model generation module, for correcting the second texture model of the character according to the multi-angle UV expansion inspection result to generate a third texture model of the character.

[0006] According to a third aspect disclosed in the present application, a storage medium is provided, on which a computer program is stored, and when the computer program is executed by a processor, any step of the first aspect disclosed in the present application is implemented.

[0007] One or more technical solutions provided in this application have at least the following technical effects or advantages:

[0008] By collecting images of the target person from multiple angles, the person's appearance information can be fully obtained. Multi-angle collection ensures the integrity and diversity of the data, providing rich input for subsequent 3D modeling and texture compensation. By detecting visible texture features of multi-angle images, the texture details of the person's surface can be accurately captured. By using the method of predetermined aiming frames and detection channels, the texture features can be regionalized and analyzed to ensure that the texture features of each part are accurately extracted. The first texture model is constructed based on the multi-angle texture feature distribution, so that the model can initially have a real texture expression. Through 3D modeling technology, the two-dimensional image features are converted into texture distribution in three-dimensional space, realizing the transformation of data form, which is used for subsequent The process of UV unfolding the second texture model at multiple angles verifies the mapping accuracy of the texture on the surface of the three-dimensional model, and determines texture errors or mapping problems through inspection, providing a clear direction for improvement for the final model optimization. The model is corrected according to the inspection results, and the problems in texture mapping are solved, so that the third texture model has higher realism and visual quality. The corrected third texture model has integrity, authenticity and consistency, which improves the overall performance of the three-dimensional character model.

[0009] The above description is only an overview of the technical solution of the present application. In order to more clearly understand the technical means of the present application, it can be implemented in accordance with the contents of the specification. In order to make the above and other purposes, features and advantages of the present application more obvious and easy to understand, the specific implementation methods of the present application are listed below. BRIEF DESCRIPTION OF THE DRAWINGS

[0010] Figure 1 A schematic flow chart of a texture generation method for a three-dimensional character model provided in an embodiment of the present application.

[0011] Figure 2 A schematic diagram of the structure of a texture generation system for a three-dimensional character model provided in an embodiment of the present application.

[0012] Explanation of the accompanying drawings: image acquisition module 10, feature detection module 20, three-dimensional modeling module 30, feature compensation module 40, expansion inspection module 50, model generation module 60. DETAILED DESCRIPTION

[0013] The embodiments of the present application provide a texture generation method, system and storage medium for a three-dimensional character model, thereby solving the technical problem in traditional three-dimensional character model generation that the limitation of the acquisition viewing angle causes the lack of texture information in some areas of the model surface, resulting in the lack of details and low authenticity of the texture of the generated model.

[0014] After introducing the basic principles of the present application, various non-limiting implementation methods of the present application will be specifically introduced below in conjunction with the accompanying drawings of the specification. It should be understood that the specific embodiments described herein are only used to explain the present application and are not used to limit the present application.

[0015] Embodiment 1, as Figure 1 As shown, the embodiment of the present application provides a texture generation method for a three-dimensional character model, the method comprising:

[0016] Multi-angle images of the target person are collected to obtain a multi-angle person image set that meets the predetermined image quality constraints.

[0017] Select the target person and ensure that the target person is in a suitable shooting position, use a high-quality camera or imaging device to shoot, ensure that the shooting equipment has high resolution, and the shooting angles include the front, side, back and other possible angles. Ensure that the image at each shooting angle meets the predetermined image quality constraints, such as clarity, color reproduction and noise-free, etc. Quality checks can be performed automatically to ensure that each image meets certain standards to avoid blurred or substandard images. Collect images from multiple angles, and each image from each angle contains a complete or partial view of the person to form a multi-angle person image set.

[0018] Visible texture feature detection is performed on the multi-angle character image set according to a predetermined aiming frame and a visible texture detection channel to obtain a multi-angle visible texture feature distribution.

[0019] The predetermined aiming frame is a rectangular or other shaped area used to define the area of ​​interest. On each image, a specific person area is cut out according to the predetermined aiming frame. By setting multiple aiming frames, different parts of the person can be captured, such as face, shoulders, hands, etc.; the visible texture detection channel includes multiple visible texture detection models, which can use deep learning-based networks, such as convolutional neural networks, or traditional image processing methods to extract texture features; visible texture features refer to texture features that can be clearly seen in the image, such as the fabric texture of clothes, skin details, hair texture, etc. These features usually have strong edges, details and local changes.

[0020] Specifically, the image is divided according to the predetermined aiming frame. For each aiming frame image, the texture richness of the area is evaluated, which can be achieved by calculating the complexity of the texture information or the local contrast. The obtained texture richness coefficient reflects the detail richness of the texture. According to each aiming frame image and its corresponding texture richness coefficient, the texture detection channel is adaptively activated, where each aiming frame image will be subjected to texture feature extraction through a series of activated models, and the texture detection results of all aiming frame images at different angles are summarized to finally generate a multi-angle visible texture feature distribution. The image at each angle will contribute certain texture features, and these feature data will serve as the basic input for 3D modeling.

[0021] Three-dimensional modeling is performed according to the multi-angle visible texture feature distribution to generate a first texture model of the character.

[0022] Using existing 3D modeling technologies, such as point cloud reconstruction and texture-based triangulated modeling algorithms, a basic 3D model is created based on the multi-angle image set of the character. This model serves as the basic framework for texture mapping, and the obtained multi-angle visible texture feature distribution is used to map the texture to the 3D model, including matching the texture features in the image with the corresponding surface parts of the model to ensure the correct position and direction of the texture. According to the surface geometric characteristics of the 3D model, the scale, direction and overlap of the texture are adjusted to avoid visual distortion. Through this step, it is ensured that each part of the model obtains the most appropriate texture effect. After completing the mapping of the basic texture, the first texture model of the character is generated, which contains all visible texture information from multiple angles.

[0023] The invisible texture features of the first texture model of the character are compensated according to the multi-angle character image set to generate a second texture model of the character.

[0024] Invisible textures refer to those textures that do not directly affect the visual effects, but still play an important role in the performance of 3D models. These textures usually affect the presentation of character models by controlling lighting, surface properties, physical properties, etc. Common invisible textures include normal maps, which increase the complexity of the model by simulating tiny details on the surface, such as wrinkles, bumps, etc., but do not change the geometric shape of the object. Although it is not directly visible, it changes the visual performance of the object by affecting the lighting effect. Similarly, it also includes roughness maps, reflectivity maps, specular highlight maps, displacement maps, transparency maps, and ambient occlusion maps.

[0025] Traverse the first texture model of the character and detect areas where texture information is missing in the model. For example, based on the frequency of color gradient changes in the texture distribution, identify areas with low texture clarity. The detected texture missing areas include completely occluded areas, such as the scalp under the character's hair, the inside of the palm, and the armpit, as well as texture transition areas, such as the boundary area when the texture is spliced ​​from one perspective to another. Use the texture data of the surrounding area and image interpolation algorithms, such as bilinear interpolation and texture extension algorithms, to fill in the texture missing areas. Compensate for the invisible texture features to correct the texture missing in the first texture model and generate a complete and accurate second texture model.

[0026] A multi-angle UV expansion test is performed on the second texture model of the character according to the multi-angle character image set to obtain a multi-angle UV expansion test result.

[0027] The second texture model is subjected to multi-angle UV unfolding, which is a process of flattening the surface of a three-dimensional model into a two-dimensional image so that mapping can be performed on the model surface. The original multi-angle character image set is used to check the effect of UV unfolding, including comparing the UV unfolding map (i.e., the two-dimensional unfolded texture map) with the corresponding parts of the actual character image, evaluating the alignment, scale and direction of the texture, and inspecting the texture clarity, realism and natural connection of the UV unfolding map from different angles, aiming to determine any inconsistencies or errors in texture mapping, such as texture breaks, duplications or blurs, etc. According to the degree of match between the unfolding map and the original image, a multi-angle UV unfolding inspection result is generated to point out specific areas that need adjustment or improvement for further optimization.

[0028] According to the multi-angle UV expansion test result, the second texture model of the character is corrected to generate a third texture model of the character.

[0029] Based on the UV unfolding test results, identify all areas that need to be adjusted, including modifications to the texture's position, scaling, rotation, or other visual effects, and make specific texture adjustments to the identified problem areas, such as using image editing software or dedicated 3D texture tools to correct texture mapping, including readjusting UV coordinates, smoothing connections, or re-mapping to eliminate visual defects. If there are significant texture problems, go back to the UV unfolding step and re-map UVs in certain areas, including adjusting the UV unfolding layout and optimizing texture mapping to accommodate complex geometric shapes. After completing all necessary adjustments, generate an optimized third texture model of the character that reflects the natural and high-quality effects of the texture when viewed from multiple angles.

[0030] Furthermore, performing visible texture feature detection on the multi-angle person image set according to a predetermined aiming frame and a visible texture detection channel to obtain a multi-angle visible texture feature distribution includes:

[0031] According to the multi-angle character image set, a first-angle character image is extracted; according to the predetermined aiming frame, the first-angle character image is divided to obtain M character aiming frame images, where M is a positive integer greater than 1; according to the M character aiming frame images, visible texture richness is evaluated to generate M visible texture richness coefficients; based on the M character aiming frame images and the M visible texture richness coefficients, according to the visible texture detection channel, M aiming frame image texture detection results are obtained; according to the M aiming frame image texture detection results, a first-angle visible texture feature distribution is generated, and the first-angle visible texture feature distribution is added to the multi-angle visible texture feature distribution.

[0032] An image at one angle is randomly selected from the multi-angle character image set as the starting point for texture feature analysis and as the first-angle character image.

[0033] Use image processing software or automated scripts to apply a predetermined aiming frame on the first-angle image to divide the image into multiple regions. Each region contains sufficient information for texture richness assessment. The area demarcated by each aiming frame becomes a separate character aiming frame image for use in subsequent steps. After division, M character aiming frame images are obtained, where M is determined based on the size of the anchor frame and the size of the original image.

[0034] Image processing techniques such as edge detection, texture filtering, and frequency domain analysis are used to analyze the texture features of each character’s aiming frame image. The visible texture richness coefficient of each aiming frame image is calculated based on factors such as texture details, contrast, and complexity. The visible texture richness coefficient reflects the richness of the texture in the image area and the complexity of the visual effect. After evaluation, a visible texture richness coefficient is generated for each aiming frame image. These coefficients are used for further texture feature integration and analysis.

[0035] Texture detection gated feature analysis is performed according to M visible texture richness coefficients, and feature activation is performed on the visible texture detection channel according to the analysis results to generate multiple activated texture detection models. Texture detection is performed on M person aiming frame images respectively using the multiple activated texture detection models to obtain multiple visible texture detection results for each aiming frame image. The multiple visible texture detection results for each aiming frame image are fused to obtain an aiming frame image texture detection result for each aiming frame image, which describes in detail various texture features identified in the image.

[0036] The texture detection results of all the M separate aiming frame images are merged, and the spatial relationship and interaction between textures are combined to create a texture feature distribution map covering the entire angle view, and the first-angle visible texture feature distribution is generated. According to the generation method of the first-angle visible texture feature distribution, the multi-angle character image set is traversed to obtain the multi-angle visible texture feature distribution.

[0037] Further, based on the M person aiming frame images and the M visible texture richness coefficients, according to the visible texture detection channel, M aiming frame image texture detection results are obtained, including:

[0038] According to the M person aiming frame images and the M visible texture rich coefficients, extract the m-th person aiming frame image and the m-th visible texture rich coefficient, wherein m is a positive integer, 1≤m≤M; perform texture detection gating feature analysis according to the m-th visible texture rich coefficient to generate the m-th texture detection gating coefficient; perform feature activation on the visible texture detection channel based on the m-th texture detection gating coefficient to generate multiple activated texture detection models, wherein the visible texture detection channel includes multiple visible texture detection models; input the m-th person aiming frame image into the multiple activated texture detection models to obtain multiple visible texture detection results; fuse the multiple visible texture detection results to generate the m-th aiming frame image texture detection result, and add the m-th aiming frame image texture detection result to the M aiming frame image texture detection results.

[0039] According to the given index m, extract the mth person aiming frame image and the corresponding mth visible texture enrichment coefficient. The index m is a positive integer ranging from 1 to M, where M is the total number of aiming frame images.

[0040] Texture detection gating feature analysis is performed based on the mth visible texture richness coefficient, that is, the coefficient is used to determine key parameters in the texture detection algorithm, such as detection sensitivity, filter threshold, etc., and the mth texture detection gating coefficient is generated according to the analysis result. This coefficient is used to adjust the operation mode of the algorithm in the texture detection algorithm, which directly affects the feature extraction and analysis strength in the texture detection process. For example, if the texture richness coefficient is high, the gating coefficient will adjust the algorithm to analyze the areas with rich details more carefully.

[0041] In the visible texture detection channel, there are multiple visible texture detection models, each model is responsible for processing different texture features, such as color, contrast, edge, detail, etc. The number of detection models to be activated is determined according to the m-th texture detection gating coefficient. For example, if the m-th texture detection gating coefficient is 4, then 4 visible texture detection models in the visible texture detection channel are activated to obtain 4 activated texture detection models.

[0042] The m-th person aiming frame image is input into multiple activated texture detection models, which independently analyze different texture features of the image. Each activated texture detection model outputs a texture detection result to obtain multiple visible texture detection results.

[0043] Normalize or standardize the output results of multiple detection models to ensure that the results of different models have the same scale and format. Use a fusion algorithm to merge multiple detection results into a comprehensive result. For example, assign weights according to the importance of each detection model and calculate the weighted average. The fused result is the texture detection result of the m-th aiming frame image, which describes the comprehensive performance of all relevant texture features in the image. Add the texture detection result of the m-th aiming frame image to the set of M aiming frame image detection results for subsequent use.

[0044] Furthermore, performing texture detection gating feature analysis according to the mth visible texture richness coefficient to generate the mth texture detection gating coefficient includes:

[0045] A texture detection gating learning model is supervised and trained with a visible texture enrichment coefficient sample set as input information and a texture detection gating coefficient sample set as output information. A texture detection gating loss coefficient is obtained after each predetermined number of trainings. It is determined whether the texture detection gating loss coefficient is less than a texture detection gating loss threshold. If the texture detection gating loss coefficient is less than the texture detection gating loss threshold, a texture detection gating channel is generated. The m-th visible texture enrichment coefficient is input into the texture detection gating channel to obtain the m-th texture detection gating coefficient.

[0046] The visible texture richness coefficient sample set includes the visible texture richness coefficient obtained by evaluating the visible texture richness in different aiming frame image samples, which serves as the input information for model training; the texture detection gating coefficient sample set is the ideal output value associated with the input data, which serves as the output information of model training, that is, the goal of model learning, and represents the target value of the gating coefficient. The data samples cover various situations, including texture areas with low richness, medium richness and high richness.

[0047] Using a regression model suitable for supervised learning tasks, such as linear regression, support vector machine, neural network, etc., the model learns how to map the visible texture enrichment coefficient to the corresponding texture detection gating coefficient. The supervised training process is as follows: the sample data is divided into a training set and a validation set, such as 80% for training and 20% for validation. The training set samples are fed into the model for supervised learning and the validation set is used for validation. The model updates the weights according to the prediction error of each iteration. After each predetermined number of training, the current loss value is recorded as the texture detection gating loss coefficient.

[0048] The texture detection gating loss threshold is a threshold pre-set according to specific needs, which represents the minimum requirement for model training. The texture detection gating loss coefficient is compared with the texture detection gating loss threshold. If the loss coefficient is less than the threshold, it means that the model prediction error is small and the performance meets the requirements. The model is judged to have reached a usable state. In this case, the trained texture detection gating learning model is deployed as a texture detection gating channel, which is used to dynamically generate the gating coefficient according to the input visible texture richness coefficient. If the loss coefficient does not meet the standard, it is necessary to readjust the model parameters, increase the training data, or optimize the training process until a gating channel that meets the requirements is generated.

[0049] The mth visible texture richness coefficient is input into the trained texture detection gating channel. The gating channel uses the trained texture detection gating learning model to calculate the mapping relationship between the visible texture richness coefficient and the gating coefficient according to the feature weights learned in training, infers the corresponding gating coefficient, and outputs the mth texture detection gating coefficient to guide the number of activated texture detection models.

[0050] Further, performing a multi-angle UV expansion test on the second texture model of the character according to the multi-angle character image set to obtain a multi-angle UV expansion test result includes:

[0051] Perform multi-angle UV unfolding according to the second texture model of the character to obtain UV unfolding images at each angle; construct UV coordinate systems at each angle according to the UV unfolding images at each angle; map the multi-angle character image set to the UV coordinate systems at each angle to obtain UV standard images at each angle; obtain the multi-texture expected factor of the target character, wherein the multi-texture expected factor includes expected texture clarity, expected texture realism and expected texture connection naturalness; based on the multi-texture expected factor, test the UV unfolding images at each angle according to the UV standard images at each angle to generate the multi-angle UV unfolding test result.

[0052] UV unfolding refers to unfolding the surface mesh of a three-dimensional model onto a two-dimensional plane to form UV texture coordinates for mapping the texture to the model surface. Specifically, according to the vertex information and texture distribution of the character's second texture model, the model surface is divided into multiple small grids, which can be triangles or quadrilaterals, so as to divide the model surface into unfoldable parts to avoid overlap or extreme stretching on the UV map. The UV coordinates of each vertex are calculated to determine the position of the three-dimensional surface point in the two-dimensional UV map. The unfolded UV coordinates are output as a UV unfolding map to display the two-dimensional texture layout of the model. By analogy, the model is unfolded from multiple angles, such as generating UV unfolding maps from different angles such as the front, side, and back. The UV unfolding map at each angle reflects the layout of the texture information in the two-dimensional plane at that perspective.

[0053] The UV coordinate system is used to describe the positional relationship in the UV unfolding diagram. The range of UV coordinates is usually [0,1]. U represents the horizontal axis and V represents the vertical axis. For UV unfolding diagrams at different angles, the corresponding UV coordinate systems are defined to form UV coordinate systems at different angles.

[0054] For each perspective of the multi-angle character image set, the corresponding image is projected onto the UV coordinate system to generate a two-dimensional texture map. The texture information in the UV space is filled through the interpolation method. Combined with the images at each angle, a UV standard map is generated. This standard map represents the real texture distribution of the model at the current perspective.

[0055] The multivariate texture expectation factor of the target person is obtained. The multivariate texture expectation factor is a plurality of standards used to measure the texture quality and the surface performance of the model. These standards ensure that the interaction between the texture and the three-dimensional model meets the requirements of reality and visual effects, including expected texture clarity, expected texture fidelity and expected texture connection naturalness. Among them, the expected texture clarity represents the details and resolution of the texture. This factor measures whether the texture has sufficient resolution and whether it can accurately represent the details in each area of ​​the model; the expected texture fidelity represents whether the texture is realistic and conforms to the physical properties of the real world. This factor measures the consistency of the texture color, light reflection, etc. with the texture effect in reality; the expected texture connection naturalness represents the smoothness and naturalness of the texture connection in different areas. This factor mainly evaluates the effects of the splicing area and the transition area to avoid obvious splicing marks or jumping feeling.

[0056] The goal of UV unfolding inspection is to evaluate the quality of UV unfolding images based on the UV standard images at each angle. The inspection criteria are multi-dimensional texture expectation factors, including texture clarity, texture fidelity, and texture connection naturalness. Specifically, an expected texture clarity comparison is performed to compare the texture resolution of each UV unfolding image with the expected texture clarity requirements to check whether there are areas with too low resolution or blurred textures; an expected texture fidelity comparison is performed to check the fidelity of color, lighting, and material for each UV unfolding image to ensure that the texture presents an effect that conforms to physical reality; an expected texture connection naturalness comparison is performed to check the naturalness of the splicing area and the transition area to ensure that there is no obvious segmentation or jump at the texture seams. By inspecting the multi-angle UV unfolding images, the multi-angle UV unfolding inspection results are generated to display the quality of the textures at each angle.

[0057] Furthermore, based on the multi-texture expected factor, the UV expansion diagrams at each angle are inspected according to the UV standard diagrams at each angle to generate the multi-angle UV expansion inspection result, including:

[0058] According to the UV standard images at each angle and the UV expansion images at each angle, a first-angle UV standard image and a first-angle UV expansion image are extracted; the texture clarity of the first-angle UV expansion image is evaluated according to the first-angle UV standard image to obtain a first texture clarity; the texture fidelity of the first-angle UV expansion image is evaluated according to the first-angle UV standard image to obtain a first texture fidelity; the texture connection naturalness of the first-angle UV expansion image is evaluated according to the first-angle UV standard image to obtain a first texture connection naturalness; the first texture clarity, the first texture fidelity and the first texture connection naturalness are used as a first-angle UV expansion detection result; it is determined whether the first-angle UV expansion detection result meets the multi-texture expected factor, a first-angle UV expansion inspection result is generated, and the first-angle UV expansion inspection result is added to the multi-angle UV expansion inspection result.

[0059] From the UV standard images at each angle and the UV expansion images at each angle, one angle is randomly selected as the first angle, and its corresponding first angle UV standard image and first angle UV expansion image are extracted. The random extraction is to evaluate the texture quality individually to ensure that the texture at each angle is fully analyzed in subsequent processing.

[0060] Texture clarity refers to the fineness of details in a texture image. The higher the clarity, the richer the details in the image and the sharper the edges of the image. The difference in clarity between the extracted first-angle UV standard image and the first-angle UV unfolded image is evaluated. Exemplarily, an edge detection algorithm, such as the Sobel operator, Canny edge detection, etc., is used to extract edge information in the image, and the texture details are evaluated by comparing the edge clarity in the standard image and the unfolded image. Based on the texture clarity analysis, a first texture clarity is generated to indicate the degree of clarity.

[0061] Texture fidelity refers to whether the texture can truly reflect the physical properties and appearance of the target object. A texture with high fidelity is consistent with the material, lighting and other characteristics of the real world. The fidelity of the texture is evaluated by comparing the color, lighting and surface characteristics of the standard image and the expanded image. For example, the lighting effects in the standard image and the expanded image are analyzed, and the consistency of the shadows, brightness and highlight areas is compared. Based on the texture fidelity analysis, the first texture fidelity is generated to indicate the realism of the texture.

[0062] The naturalness of texture connection refers to whether the transition between different areas of the texture is natural and smooth. This evaluation standard focuses on whether the texture seams and transition areas are seamlessly connected to avoid obvious splicing marks. It evaluates the natural transition degree of texture seams and splicing in the first-angle UV expansion image. For example, the seam area of ​​the standard image and the expansion image is compared to analyze whether the color, lighting, texture pattern, etc. of the splicing are smoothly transitioned; through features such as texture gradient, lighting inconsistency and color change, detect whether there are obvious seams, breaks or unnatural edges. Based on the texture connection naturalness analysis, the first texture connection naturalness is generated to represent the naturalness of texture connection.

[0063] The first texture clarity, the first texture fidelity and the first texture connection naturalness are integrated as the first angle UV unfolding detection result, which will determine whether the texture quality at this angle meets expectations.

[0064] The first texture clarity, the first texture fidelity and the first texture connection naturalness in the first angle UV expansion detection result are compared with the expected texture clarity, the expected texture fidelity and the expected texture connection naturalness of the multivariate texture expected factor, respectively. If the first angle UV expansion detection result meets the expected factors, the first angle UV expansion inspection result is generated as passed, indicating that the UV expansion of this angle meets the quality requirements; conversely, if the first angle UV expansion detection result does not meet the expected factors, the first angle UV expansion inspection result is generated as failed, indicating that the UV expansion of this angle does not meet the quality requirements.

[0065] The first-angle UV unfolding test results (pass / fail) are added to the multi-angle UV unfolding test results to build a complete set of test results. The test results of each angle will help judge the quality of the overall texture model and facilitate subsequent optimization and correction.

[0066] Furthermore, according to the multi-angle UV expansion test result, the second texture model of the character is corrected to generate a third texture model of the character, including:

[0067] If any angle UV expansion test result in the multi-angle UV expansion test results fails, an abnormal UV expansion test result is generated; UV expansion test backtracking is performed according to the abnormal UV expansion test result to determine the UV expansion test abnormality factor; based on the UV expansion test abnormality factor, according to the angle UV standard map corresponding to the abnormal UV expansion test result, texture deviation detection is performed on the angle UV expansion map corresponding to the abnormal UV expansion test result to obtain a texture abnormal deviation detection result; texture feature correction is performed on the second texture model of the character according to the texture abnormal deviation detection result to obtain the third texture model of the character.

[0068] When the UV unfolding test result of any angle in the multi-angle UV unfolding test results fails, it means that the UV unfolding of this angle does not meet the quality requirements, and there may be problems with texture clarity, realism or natural connection, and it will be marked as an abnormal UV unfolding test result.

[0069] Conduct retrospective analysis on abnormal UV unfolding test results to find out the root cause of failure. Specifically, for failed angle tests, analyze the various texture quality indicators of the corresponding angle to confirm which indicator does not meet expectations, including texture clarity, texture fidelity, and texture connection naturalness. For example, the abnormal UV unfolding test result is the first angle UV unfolding test result. At this time, the first texture clarity does not meet the expected texture clarity. Then, the UV unfolding test abnormal factor is the first texture clarity. Subsequently, deviation detection and compensation of the first texture clarity are required.

[0070] Based on the UV expansion inspection abnormality factor, first locate the corresponding angle UV expansion image and angle UV standard image, and determine the deviation by comparing the difference between the two. The deviation area can be detected according to the contrast, color distribution, texture pattern and other attributes of the area, with special attention paid to texture boundaries and seam areas. Common deviation types include texture stretching or compression, color difference, inconsistent lighting, discontinuous splicing, etc. Through deviation detection, a result including the deviation type and detection area is generated, that is, the texture abnormality deviation detection result, which is used to point out the specific location of the problem, such as texture stretching area, color deviation area, etc.

[0071] According to the texture abnormal deviation detection results, the second texture model of the character is texture corrected to correct the abnormal areas in the texture. For example, for texture stretching or compression, the texture stretching or compression problem is corrected by texture remapping or UV coordinate adjustment to restore it to normal proportion; for color deviation, a color correction algorithm such as color mapping, histogram equalization, etc. is used to adjust the color of the texture to make it consistent with the color in the standard image to ensure a natural color transition. After the above correction, a higher quality third texture model of the character is generated. The texture quality of this model is optimized, and all deviations are corrected to ensure that the texture performance at different angles meets the multi-texture expectation factor, with better clarity, realism and natural connection.

[0072] In summary, the texture generation method of the three-dimensional character model provided in the embodiment of the present application has the following technical effects:

[0073] By collecting images of the target person from multiple angles, the person's appearance information can be fully obtained. Multi-angle collection ensures the integrity and diversity of the data, providing rich input for subsequent 3D modeling and texture compensation. By detecting visible texture features in multi-angle images, the texture details of the person's surface can be accurately captured. By using the method of predetermined aiming frames and detection channels, the texture features can be regionalized for analysis to ensure that the texture features of each part are accurately extracted. The first texture model is constructed based on the multi-angle texture feature distribution, so that the model can initially have a real texture expression. Through 3D modeling technology, the two-dimensional image features are converted into texture distribution in three-dimensional space, realizing the transformation of data form, which is used for subsequent The process of UV unfolding the second texture model at multiple angles verifies the mapping accuracy of the texture on the surface of the three-dimensional model, and determines texture errors or mapping problems through inspection, providing a clear direction for improvement for the final model optimization. The model is corrected according to the inspection results, and the problems in texture mapping are solved, so that the third texture model has higher realism and visual quality. The corrected third texture model has integrity, authenticity and consistency, which improves the overall performance of the three-dimensional character model.

[0074] Embodiment 2, based on the same inventive concept as the texture generation method of the three-dimensional character model in the above embodiment, Figure 2 As shown, an embodiment of the present application provides a texture generation system for a three-dimensional character model, the system comprising:

[0075] The image acquisition module 10 is used to perform multi-angle image acquisition on the target person to obtain a multi-angle person image set that meets the predetermined image quality constraints; the feature detection module 20 is used to perform visible texture feature detection on the multi-angle person image set according to the predetermined aiming frame and the visible texture detection channel to obtain a multi-angle visible texture feature distribution; the three-dimensional modeling module 30 is used to perform three-dimensional modeling according to the multi-angle visible texture feature distribution to generate a first texture model of the person; the feature compensation module 40 is used to perform invisible texture feature compensation on the first texture model of the person according to the multi-angle person image set to generate a second texture model of the person; the expansion inspection module 50 is used to perform multi-angle UV expansion inspection on the second texture model of the person according to the multi-angle person image set to obtain a multi-angle UV expansion inspection result;

[0076] The model generation module 60 is used to correct the second texture model of the character and generate a third texture model of the character according to the multi-angle UV expansion test result.

[0077] Furthermore, the feature detection module 20 includes the following operation steps:

[0078] According to the multi-angle character image set, a first-angle character image is extracted; according to the predetermined aiming frame, the first-angle character image is divided to obtain M character aiming frame images, where M is a positive integer greater than 1; according to the M character aiming frame images, visible texture richness is evaluated to generate M visible texture richness coefficients; based on the M character aiming frame images and the M visible texture richness coefficients, according to the visible texture detection channel, M aiming frame image texture detection results are obtained; according to the M aiming frame image texture detection results, a first-angle visible texture feature distribution is generated, and the first-angle visible texture feature distribution is added to the multi-angle visible texture feature distribution.

[0079] Furthermore, the feature detection module 20 includes the following operation steps:

[0080] According to the M person aiming frame images and the M visible texture rich coefficients, extract the m-th person aiming frame image and the m-th visible texture rich coefficient, wherein m is a positive integer, 1≤m≤M; perform texture detection gating feature analysis according to the m-th visible texture rich coefficient to generate the m-th texture detection gating coefficient; perform feature activation on the visible texture detection channel based on the m-th texture detection gating coefficient to generate multiple activated texture detection models, wherein the visible texture detection channel includes multiple visible texture detection models; input the m-th person aiming frame image into the multiple activated texture detection models to obtain multiple visible texture detection results; fuse the multiple visible texture detection results to generate the m-th aiming frame image texture detection result, and add the m-th aiming frame image texture detection result to the M aiming frame image texture detection results.

[0081] Furthermore, the feature detection module 20 includes the following operation steps:

[0082] A texture detection gating learning model is supervised and trained with a visible texture enrichment coefficient sample set as input information and a texture detection gating coefficient sample set as output information. A texture detection gating loss coefficient is obtained after each predetermined number of trainings. It is determined whether the texture detection gating loss coefficient is less than a texture detection gating loss threshold. If the texture detection gating loss coefficient is less than the texture detection gating loss threshold, a texture detection gating channel is generated. The m-th visible texture enrichment coefficient is input into the texture detection gating channel to obtain the m-th texture detection gating coefficient.

[0083] Furthermore, the deployment inspection module 50 includes the following operation steps:

[0084] Perform multi-angle UV unfolding according to the second texture model of the character to obtain UV unfolding images at each angle; construct UV coordinate systems at each angle according to the UV unfolding images at each angle; map the multi-angle character image set to the UV coordinate systems at each angle to obtain UV standard images at each angle; obtain the multi-texture expected factor of the target character, wherein the multi-texture expected factor includes expected texture clarity, expected texture realism and expected texture connection naturalness; based on the multi-texture expected factor, test the UV unfolding images at each angle according to the UV standard images at each angle to generate the multi-angle UV unfolding test result.

[0085] Furthermore, the deployment inspection module 50 includes the following operation steps:

[0086] According to the UV standard images at each angle and the UV expansion images at each angle, a first-angle UV standard image and a first-angle UV expansion image are extracted; the texture clarity of the first-angle UV expansion image is evaluated according to the first-angle UV standard image to obtain a first texture clarity; the texture fidelity of the first-angle UV expansion image is evaluated according to the first-angle UV standard image to obtain a first texture fidelity; the texture connection naturalness of the first-angle UV expansion image is evaluated according to the first-angle UV standard image to obtain a first texture connection naturalness; the first texture clarity, the first texture fidelity and the first texture connection naturalness are used as a first-angle UV expansion detection result; it is determined whether the first-angle UV expansion detection result meets the multi-texture expected factor, a first-angle UV expansion inspection result is generated, and the first-angle UV expansion inspection result is added to the multi-angle UV expansion inspection result.

[0087] Furthermore, the model generation module 60 includes the following steps:

[0088] If any angle UV expansion test result in the multi-angle UV expansion test results fails, an abnormal UV expansion test result is generated; UV expansion test backtracking is performed according to the abnormal UV expansion test result to determine the UV expansion test abnormality factor; based on the UV expansion test abnormality factor, according to the angle UV standard map corresponding to the abnormal UV expansion test result, texture deviation detection is performed on the angle UV expansion map corresponding to the abnormal UV expansion test result to obtain a texture abnormal deviation detection result; texture feature correction is performed on the second texture model of the character according to the texture abnormal deviation detection result to obtain the third texture model of the character.

[0089] Through the above-mentioned detailed description of the texture generation method of the three-dimensional character model in this specification, those skilled in the art can clearly understand the texture generation system of the three-dimensional character model in this embodiment. Since it corresponds to the method disclosed in the embodiment, the description is relatively simple, and the relevant parts can be referred to the method part description.

[0090] Embodiment 3, a storage medium having a computer program stored thereon, wherein the computer program, when executed by a processor, implements the texture generation method for a three-dimensional character model of the aforementioned embodiment 1.

[0091] The above description of the disclosed embodiments enables those skilled in the art to implement or use the present application. Various modifications to these embodiments will be apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the present application. Therefore, the present application will not be limited to the embodiments shown herein, but will conform to the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A texture generation method for a three-dimensional character model, characterized in that: The method comprises: Capturing multi-angle images of the target person to obtain a multi-angle person image set that meets a predetermined image quality constraint; Performing visible texture feature detection on the multi-angle person image set according to a predetermined aiming frame and a visible texture detection channel to obtain a multi-angle visible texture feature distribution; Perform three-dimensional modeling according to the multi-angle visible texture feature distribution to generate a first texture model of the character; Perform invisible texture feature compensation on the first texture model of the character according to the multi-angle character image set to generate a second texture model of the character; Perform a multi-angle UV unfolding test on the second texture model of the character according to the multi-angle character image set to obtain a multi-angle UV unfolding test result; Correcting the second texture model of the character according to the multi-angle UV expansion test result to generate a third texture model of the character; Among them, performing visible texture feature detection on the multi-angle character image set according to the predetermined aiming frame and the visible texture detection channel to obtain the multi-angle visible texture feature distribution includes: Extracting a first-angle character image according to the multi-angle character image set; Dividing the first-angle character image according to the predetermined aiming frame to obtain M character aiming frame images, where M is a positive integer greater than 1; Performing visible texture richness evaluation based on the M person aiming frame images to generate M visible texture richness coefficients; Based on the M person aiming frame images and the M visible texture richness coefficients, and according to the visible texture detection channel, M aiming frame image texture detection results are obtained; A first-angle visible texture feature distribution is generated according to the texture detection results of the M aiming frame images, and the first-angle visible texture feature distribution is added to the multi-angle visible texture feature distribution.

2. The method according to claim 1, characterized in that Based on the M person aiming frame images and the M visible texture richness coefficients, and according to the visible texture detection channel, M aiming frame image texture detection results are obtained, including: Extracting the mth person aiming frame image and the mth visible texture rich coefficient according to the M person aiming frame images and the M visible texture rich coefficients, wherein m is a positive integer, 1≤m≤M; Performing texture detection gating feature analysis according to the mth visible texture richness coefficient to generate an mth texture detection gating coefficient; Performing feature activation on the visible texture detection channel based on the mth texture detection gating coefficient to generate a plurality of activated texture detection models, wherein the visible texture detection channel includes a plurality of visible texture detection models; Inputting the m-th person aiming frame image into the multiple activated texture detection models to obtain multiple visible texture detection results; The multiple visible texture detection results are fused to generate an m-th aiming frame image texture detection result, and the m-th aiming frame image texture detection result is added to the M aiming frame image texture detection results.

3. The method according to claim 2, characterized in that Performing texture detection gating feature analysis according to the mth visible texture richness coefficient to generate the mth texture detection gating coefficient includes: Using a visible texture richness coefficient sample set as input information and a texture detection gating coefficient sample set as output information, the texture detection gating learning model is supervised and trained, and a texture detection gating loss coefficient is obtained after each predetermined number of trainings; Determining whether the texture detection gating loss coefficient is less than a texture detection gating loss threshold; If the texture detection gating loss coefficient is less than the texture detection gating loss threshold, generating a texture detection gating channel; The mth visible texture richness coefficient is input into the texture detection gating channel to obtain the mth texture detection gating coefficient.

4. The method according to claim 1, characterized in that Performing a multi-angle UV expansion test on the second texture model of the character according to the multi-angle character image set to obtain a multi-angle UV expansion test result, including: Perform multi-angle UV unfolding according to the second texture model of the character to obtain UV unfolding images at each angle; Constructing UV coordinate systems at various angles according to the UV expansion diagrams at various angles; Mapping the multi-angle character image set to the UV coordinate system of each angle to obtain a UV standard image of each angle; Obtaining a multi-texture expected factor of the target person, wherein the multi-texture expected factor includes expected texture clarity, expected texture fidelity, and expected texture connection naturalness; Based on the multi-texture expected factor, the UV expansion diagrams at each angle are inspected according to the UV standard diagrams at each angle to generate the multi-angle UV expansion inspection result.

5. The method according to claim 4, characterized in that Based on the multi-texture expected factor, the UV expansion diagrams at each angle are inspected according to the UV standard diagrams at each angle to generate the multi-angle UV expansion inspection result, including: Extracting a first-angle UV standard image and a first-angle UV expansion image according to the UV standard images at each angle and the UV expansion images at each angle; Performing a texture clarity evaluation on the first angle UV expansion image according to the first angle UV standard image to obtain a first texture clarity; Performing texture fidelity evaluation on the first angle UV expansion image according to the first angle UV standard image to obtain a first texture fidelity; According to the first angle UV standard image, the texture connection naturalness of the first angle UV expansion image is evaluated to obtain a first texture connection naturalness; Taking the first texture clarity, the first texture fidelity and the first texture connection naturalness as the first angle UV unfolding detection result; It is determined whether the first-angle UV unfolding detection result satisfies the multi-texture expectation factor, a first-angle UV unfolding inspection result is generated, and the first-angle UV unfolding inspection result is added to the multi-angle UV unfolding inspection result.

6. The method according to claim 1, characterized in that According to the multi-angle UV expansion test result, the second texture model of the character is corrected to generate a third texture model of the character, including: If any angle UV unfolding test result in the multi-angle UV unfolding test results fails, an abnormal UV unfolding test result is generated; Perform UV inspection backtracking according to the abnormal UV inspection result to determine the UV inspection abnormality factor; Based on the UV expansion test abnormality factor, and according to the angle UV standard map corresponding to the abnormal UV expansion test result, a texture deviation detection is performed on the angle UV expansion map corresponding to the abnormal UV expansion test result to obtain a texture abnormal deviation detection result; The texture feature correction is performed on the second texture model of the character according to the texture abnormality deviation detection result to obtain the third texture model of the character.

7. A texture generation system for a three-dimensional character model, characterized in that: A method for generating a texture of a three-dimensional character model according to any one of claims 1 to 6, the system comprising: An image acquisition module is used to acquire multi-angle images of a target person to obtain a multi-angle person image set that meets a predetermined image quality constraint; A feature detection module, used to perform visible texture feature detection on the multi-angle person image set according to a predetermined aiming frame and a visible texture detection channel to obtain a multi-angle visible texture feature distribution; A three-dimensional modeling module, used for performing three-dimensional modeling according to the multi-angle visible texture feature distribution to generate a first texture model of the character; A feature compensation module, used for performing invisible texture feature compensation on the first texture model of the character according to the multi-angle character image set to generate a second texture model of the character; An unfolding inspection module, used for performing a multi-angle UV unfolding inspection on the second texture model of the character according to the multi-angle character image set to obtain a multi-angle UV unfolding inspection result; The model generation module is used to correct the second texture model of the character and generate a third texture model of the character according to the multi-angle UV expansion test result.

8. A storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the texture generation method for a three-dimensional character model according to any one of claims 1 to 6 are implemented.

Images