A texture map generation method and device based on physiological parameter decomposition

Abstract

The present invention discloses a method and device for generating texture maps based on physiological parameter decomposition, including: downsampling an ultra-high-resolution portrait and passing it through a texture mapping predictor to generate a global texture mapping feature, the global texture mapping feature is position-encoded to obtain a position-encoded feature and then passed through a physiological parameter predictor to generate a multi-channel physiological parameter texture, the physiological parameter texture is mapped to a global albedo by a mapper; the ultra-high-resolution portrait is divided into a plurality of portrait small blocks and then input into a generator, each portrait small block generates a local texture small block under the guidance of the global albedo, and all the local texture small blocks are assembled and mapped to a geometric body to obtain a final texture map. In this way, the skin albedo in the reconstructed texture map is more stable and natural, and can maintain accuracy in extreme lighting scenarios.

Description

Technical Field

[0001] The invention belongs to the technical field of the combination of computer graphics and image processing, and in particular relates to a texture map generation method and device based on physiological parameter decomposition. Background Art

[0002] With the continuous development of computer performance, the academic and industrial circles have an increasing demand for faster and higher-quality rendering of realistic images in computers. Up to now, graphics has made breakthroughs in many fields, such as physically based rendering, inverse rendering, 3D reconstruction, real-time acceleration, etc., but in the direction of rendering simulation of human facial skin, there have been few breakthroughs in recent years. On the one hand, the surface-based skin rendering model is sufficient to meet the needs in ordinary games or film and television scenes after combining some heuristic priors. On the other hand, rendering human skin effects full of details and correct skin transmission requires complex and high-precision materials, such as micro-normal maps that represent the micro-surface of facial pores, and high-resolution albedo maps that accurately represent facial chromophores. One point that makes skin rendering more difficult than other traditional object rendering is that a slight reduction in the accuracy of skin rendering may lead to a significant decline in sensory experience. Therefore, studying how to faithfully obtain high-resolution and physiologically reasonable texture maps from real-life images has become an important research direction.

[0003] At present, there are many high-resolution face reconstruction methods in academia, such as UltraAvatar [A RealisticAnimatable 3D Avatar Diffusion Model with Authenticity Guided Textures, a realistic animatable 3D avatar diffusion model with realistic guided textures] and DreamFace [DreamFace: ProgressiveGeneration of Animatable 3D Faces under Text Guidance, which can dynamically generate three-dimensional faces under text guidance], which support the use of text and images as input to generate geometric and material information. In particular, DreamFace supports the use of 4K input images to generate diffuse, micronormal and specular maps with a maximum resolution of 4K. However, these methods do not explicitly introduce skin physiological parameters during reconstruction, so that the reconstructed skin albedo is not constrained by physiological parameters, which may lead to the reconstruction of skin colors that do not exist in nature in some complex lighting conditions. Summary of the invention

[0004] In view of the above, an object of the present invention is to provide a method and device for generating a texture map based on physiological parameter decomposition, so that the skin albedo in the reconstructed texture map is more stable and natural, and can maintain accuracy under extreme lighting scenarios.

[0005] To achieve the above-mentioned purpose of the invention, an embodiment provides a method for generating a texture map based on physiological parameter decomposition, comprising the following steps:

[0006] Global stage: After downsampling the ultra-high-resolution portrait, the global texture mapping feature is generated by the texture mapping predictor. The global texture mapping feature is position-encoded to obtain the position encoding feature, and then passed through the physiological parameter predictor to generate a multi-channel physiological parameter texture. The physiological parameter texture is mapped to the global albedo by the mapper;

[0007] Local stage: The ultra-high-resolution portrait is divided into multiple portrait blocks and then input into the generator. Each portrait block generates a local texture block under the guidance of the global albedo, and all local texture blocks are assembled and mapped to the geometry to obtain the final texture map.

[0008] The present invention is based on a two-stage global plus local prediction pipeline. By adding a physiological parameter predictor and a mapper, the predicted global albedo of the reaction skin color is constrained within the range that the physiological parameter predictor can represent, making the albedo prediction process more stable under different extreme lighting scenarios, and the predicted physiological parameter texture can be guaranteed to be within a reasonable range. At the same time, due to the use of a divide-and-conquer two-stage global plus local structure, high-resolution images up to 4K (i.e., 4096 pixels × 4096 pixels) can be processed end-to-end.

[0009] Preferably, the physiological parameter predictor is constructed based on a neural network, and the multi-channel physiological parameter texture generated by the predictor includes melanin distribution, melanin type mixing ratio, and hemoglobin distribution.

[0010] Preferably, when constructing a physiological parameter predictor based on a neural network, the loss function for parameter optimization includes content loss and smoothness loss, wherein the content loss is constructed based on the difference between the generated physiological parameter texture and the real physiological parameter texture, and the smoothness loss is constructed based on the horizontal gradient and vertical gradient of the generated physiological parameter texture.

[0011] Preferably, the mapper is constructed based on a neural network, which is used to map the global albedo of the skin based on the physiological parameter texture, and the construction process is:

[0012] First, according to formula (1) and formula (2), the lookup table of different physiological parameter value combinations to the epidermal absorption coefficient and the dermal absorption coefficient is calculated, and then the Monte Carlo random walk algorithm is used to calculate the albedo value under different absorption coefficients, thereby forming a mapping lookup table of different physiological parameter value combinations to albedo;

[0013] (1);

[0014] (2);

[0015] in, is the light spectrum distribution, represents the absorption coefficient of eumelanin, represents the absorption coefficient of melanin, represents other absorption coefficients, represents the oxygenated hemoglobin absorption coefficient, represents the absorption coefficient of deoxyhemoglobin, represents the bilirubin absorption coefficient, Represents the distribution of melanin, Represents the ratio of melanin type mixture, represents the distribution of hemoglobin, represents the epidermal absorption coefficient, represents the dermis absorption coefficient, Represents the proportion of oxygenated hemoglobin in hemoglobin;

[0016] Finally, the mapping lookup table is used as sample data to train the neural network to learn the mapping relationship between different physiological parameter value combinations and albedo to obtain the mapper.

[0017] Preferably, the generator adopts a diffusion neural network. During the generation process, each portrait block is used as an initial vector, and in each time step of inverse diffusion, the global albedo of the corresponding position of the portrait block is used as a guiding condition and denoising noise is generated based on the neural network, and denoising is performed based on the denoising noise to obtain a local texture block.

[0018] Preferably, the texture mapping predictor is constructed based on a neural network, the mapper is constructed based on a U-net network, and the physiological parameter predictor is constructed based on a diffusion neural network.

[0019] Preferably, the method further comprises: editing the generated multi-channel physiological parameter texture to adjust the skin color, and mapping the edited physiological parameter texture into the global albedo of different skin colors through a mapper. In this way, the reconstructed global albedo of different skin colors can be correctly modified from the perspective of physiological representation by controlling the physiological parameter texture.

[0020] To achieve the above-mentioned purpose of the invention, an embodiment of the present invention further provides a texture map generation device based on physiological parameter decomposition, comprising:

[0021] A global module is used to downsample the ultra-high-resolution portrait and generate global texture mapping features through a texture mapping predictor. The global texture mapping features are position-encoded to obtain position encoding features and then passed through a physiological parameter predictor to generate multi-channel physiological parameter textures. The physiological parameter textures are mapped to global albedo through a mapper.

[0022] The local module is used to divide the ultra-high-resolution portrait into multiple portrait blocks and input them into the generator. Each portrait block generates a local texture block under the guidance of the global albedo, and all local texture blocks are assembled and mapped to the geometry to obtain the final texture map.

[0023] To achieve the above-mentioned purpose of the invention, an embodiment further provides a computing device, including a memory and one or more processors, wherein the memory stores executable code, and when the one or more processors execute the executable code, they are used to implement the above-mentioned texture map generation method based on physiological parameter decomposition.

[0024] To achieve the above-mentioned purpose of the invention, an embodiment further provides a computer-readable storage medium on which a program is stored. When the program is executed by a processor, the above-mentioned texture map generation method based on physiological parameter decomposition is implemented.

[0025] The present invention can significantly improve the accuracy and stability of high-resolution facial texture reconstruction by introducing physiological parameter textures representing skin color and combining a global and local two-stage prediction pipeline. Compared with the prior art, the present invention has at least the following beneficial effects:

[0026] 1. Improve the naturalness and stability of the reconstructed portrait texture map: By constraining the albedo that reflects the skin color with physiological parameters, the unnatural skin color that may appear in complex lighting scenarios in the existing technology is avoided. The reconstruction result maintains a higher sense of reality under various lighting conditions and conforms to the physiological characteristics of the skin in nature.

[0027] 2. Enhanced detail retention capability: Due to the use of high-resolution images (supporting up to 4K input) and the constraints of physiological parameters, the fine features of facial skin, such as pores and wrinkles, can be captured more accurately. The generated texture patches are more delicate and realistic, avoiding the problem of detail loss in traditional methods.

[0028] 3. Improve the controllability of rendering effects: Through the control of physiological parameters, the albedo can be modified physiologically correctly while ensuring physiological rationality, providing more flexible customization options to meet different virtual image requirements.

[0029] 4. Reduce the need for manual correction: Since the present invention can automatically and accurately generate a global albedo that conforms to physiological characteristics, it reduces the need for manual intervention, improves production efficiency, and reduces the cost of later corrections.

[0030] 5. Applicable to a wide range of application scenarios: The present invention is not only suitable for the creation of high-resolution virtual images, but can also be widely used in virtual reality, digital film and television production, game development and other fields to meet the demand for high-quality facial rendering. BRIEF DESCRIPTION OF THE DRAWINGS

[0031] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.

[0032] Figure 1 is a flow chart of a method for generating a texture map based on physiological parameter decomposition provided by an embodiment;

[0033] Figure 2 is a block flow chart of texture map generation based on physiological parameter decomposition provided by an embodiment;

[0034] Figure 3 It is a structural schematic diagram of a texture map generation device based on physiological parameter decomposition provided in an embodiment. DETAILED DESCRIPTION

[0035] To make the purpose, technical solution and advantages of the present invention more clearly understood, the present invention is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific implementation methods described herein are only used to explain the present invention and do not limit the scope of protection of the present invention.

[0036] The inventive concept of the present invention is that the existing methods do not explicitly consider the deficiency of skin physiological parameters in high-resolution facial material reconstruction. Specifically, when processing high-resolution facial images, there is a lack of constraints on physiological parameters, resulting in the generated skin albedo may not conform to the physiological characteristics in nature, especially under complex lighting conditions, unnatural skin color may appear in the reconstructed texture map. To solve this technical problem, the embodiment of the present invention provides a texture map generation method and device based on physiological parameter decomposition, which physiologically and reasonably modifies the skin albedo texture of the face by introducing and controlling skin physiological parameters, so that the skin albedo in the reconstructed texture is more stable and natural, and can maintain accuracy in extreme lighting scenarios, while also solving the technical problem that the existing method directly modifies the skin texture may be physiologically incorrect.

[0037] like Figure 1 and Figure 2 As shown, the embodiment provides a method for generating a texture map based on physiological parameter decomposition, comprising the following steps:

[0038] S1, global stage: the ultra-high resolution portrait is downsampled and then passed through the texture mapping predictor to generate global texture mapping features. The global texture mapping features are position encoded to obtain position encoding features and then passed through the physiological parameter predictor to generate multi-channel physiological parameter textures. The physiological parameter textures are mapped to global albedo by the mapper.

[0039] In an embodiment, the global stage is used to extract a global albedo from an ultra-high resolution portrait for guiding skin color, and specifically, an ultra-high resolution portrait (e.g., 4K (4096×4096 pixels) is downsampled (e.g., downsampled to 768×768 pixels) and input into a texture mapping predictor to generate global texture mapping features, wherein the texture mapping predictor is constructed based on a neural network and can learn mapping features between an input image and a texture therein.

[0040] The obtained global texture mapping features are positionally encoded to obtain position encoding features. Specifically, a position encoding network is used for position encoding. The position encoding network is composed of basic structures such as convolutional layers and fully connected layers, and outputs channel position encoding features.

[0041] The albedo of human skin can be abstracted into a binary function of hemoglobin and melanin distribution. Therefore, in theory, the skin appearance effect in most cases can be simulated by controlling its distribution. By making good use of this physiological characteristic as a priori, the accuracy of the reconstructed skin albedo can be improved. Based on this, the present invention introduces a physiological parameter predictor, and the obtained position coding features are input into the physiological parameter predictor to generate a multi-channel physiological parameter texture, wherein the physiological parameter texture includes melanin distribution, melanin type mixing ratio, and hemoglobin distribution. These physiological parameter textures are mapped by a mapper to reflect the global albedo of skin color.

[0042] In an embodiment, a mapper is configured to map the physiological parameter texture to an albedo.

[0043] The physiological parameter to albedo mapper is a pre-trained module, which can use a neural network, specifically a neural network structure composed of UNet, and its function is to calculate the mapping of the three physiological parameter textures predicted by the physiological parameter predictor to the final facial skin albedo. The specific construction process of the mapper is:

[0044] First, according to formula (1) and formula (2), the lookup table of different physiological parameter value combinations to the epidermal absorption coefficient and the dermal absorption coefficient is calculated, and then the Monte Carlo random walk algorithm is used to calculate the albedo value under different absorption coefficients, thereby forming a mapping lookup table of different physiological parameter value combinations to albedo;

[0045] (1);

[0046] (2);

[0047] in, is the light spectrum distribution, represents the absorption coefficient of eumelanin, represents the absorption coefficient of melanin, represents other absorption coefficients, represents the oxygenated hemoglobin absorption coefficient, represents the absorption coefficient of deoxyhemoglobin, represents the bilirubin absorption coefficient, Represents the distribution of melanin, Represents the ratio of melanin type mixture, represents the distribution of hemoglobin, represents the epidermal absorption coefficient, represents the dermis absorption coefficient, Represents the proportion of oxygenated hemoglobin in hemoglobin;

[0048] Finally, the mapping lookup table is used as sample data to train the neural network to learn the mapping relationship between different physiological parameter value combinations and albedo, so as to obtain a mapper. Specifically, based on the mapping lookup table, different physiological parameter value combinations are used as input, and the corresponding albedo is used as output to supervise the neural network so that the neural network learns the mapping relationship in the mapping lookup table to obtain a mapper.

[0049] In the embodiment, the physiological parameter predictor is constructed based on a neural network, and specifically a diffusion neural network based on the UNet architecture can be used, for example, a pre-trained Stable Diffusion 2.1 model (stable diffusion model version 2.1) is used. The physiological parameter predictor outputs three 768×768 physiological parameter textures based on the position encoding features of the input size of 768×768×8. During training, the paired training set from super-resolution portrait to albedo is known, and combined with the mapping relationship of the previously constructed physiological parameter texture combination to albedo, the real physiological parameter texture corresponding to the super-resolution portrait can be obtained, that is, the paired training set of super-resolution portrait to real physiological parameter is obtained, and the training set is used to train the physiological parameter mapper. The specific process is:

[0050] (1) Initialize the network: Use the pre-trained Stable Diffusion 2.1 model to initialize the network. The network has been trained on a large-scale image dataset and has excellent generation capabilities. Assume that the generation process of the network is:

[0051] ;

[0052] Among them, z is the input position encoding feature, 𝐺 is the Stable Diffusion2.1 network, 𝜃 is the network parameter, and 𝑦 is the generated physiological parameter texture map.

[0053] (2) Pre-training process: Use the following loss function To optimize:

[0054] ;

[0055] in, represents the content loss, represents the smoothing loss, Represents the weighted coefficient corresponding to the smooth loss, specifically:

[0056] Content loss It is constructed based on the difference between the generated physiological parameter texture and the real physiological parameter texture. Specifically, the error of the texture content is minimized by comparing the pixel difference between the generated physiological parameter texture and the real physiological parameter texture. The calculation formula is:

[0057] ;

[0058] in, and Represents the generated physiological parameter texture and the real physiological parameter texture. i pixel values, N is the total number of pixels;

[0059] Smoothing loss By controlling the smoothness of the generated physiological parameter texture, unnatural noise or excessive details can be avoided. The calculation formula is:

[0060] ;

[0061] Among them, ∇x and ∇y are the gradient operations of the physiological parameter texture map in the horizontal and vertical directions, respectively;

[0062] (3) Optimization process: Optimizing network parameters through back propagation algorithm , the goal is to minimize the loss function .

[0063] It should be noted that the position encoding network is jointly trained with the physiological parameter predictor so that the gradient from the physiological parameter predictor can optimize the parameters of the position encoding network. It is worth mentioning that in the global stage, the mapper from physiological parameters to albedo does not participate in the training, but only plays the role of parameter mapping and gradient conduction.

[0064] In an embodiment, the user may also edit the physiological parameter texture output by the physiological parameter predictor to obtain a global albedo that is different but still complies with physiological constraints, and the obtained global albedo is used to guide local texture generation.

[0065] S2, local stage: The ultra-high-resolution portrait is divided into multiple portrait blocks and then input into the generator. Each portrait block generates a local texture block under the guidance of the global albedo, and all local texture blocks are assembled and mapped to the geometry to obtain the final texture map.

[0066] In order to obtain a high-resolution texture with local details, predictions need to be made at the original resolution level. In this stage, each portrait patch is processed separately, while the global albedo generated in the global stage is used as a guide to ensure overall consistency. Specifically, the input super-resolution portrait is first cropped into 8×8 patches, and the size of each portrait patch is (512x512 pixels). After that, each portrait patch is fed into the generator, and the texture at the corresponding position from the global albedo is used as a conditional control to guide the generation of the result. The generated local textures include albedo, micronormal, and specular map patches. Afterwards, these local textures are assembled into a complete texture of size 4096×4096 pixels. The assembly process is as follows:

[0067] ;

[0068] in, Indicates coordinate position The texture value of represents the remainder operation, P represents the texture operation, and the symbol Indicates rounding.

[0069] After the conversion, the entire texture is unfolded on the specified geometry to obtain the final texture map with a resolution of 4096×4096, that is, the final albedo, micronormal and specular map texture.

[0070] In the embodiment, the generator uses a diffusion neural network. During the generation process, each portrait block is used as an initial vector, and in each time step of the reverse diffusion, the global albedo of the corresponding position of the portrait block is used as a guiding condition and a denoising noise is generated based on the neural network, and denoising is performed based on the denoising noise to obtain a local texture block. When optimizing the parameters of the generator, the loss function of the generator is consistent with the loss function of the physiological parameter predictor.

[0071] like Figure 3 As shown, the embodiment also provides a texture map generating device 30 based on physiological parameter decomposition, including a global module 31 and a local module 32, wherein the global module 31 is used to generate a global texture mapping feature after downsampling the ultra-high resolution portrait through a texture mapping predictor, the global texture mapping feature is position-encoded to obtain a position encoding feature and then passed through a physiological parameter predictor to generate a multi-channel physiological parameter texture, and the physiological parameter texture is mapped to a global albedo through a mapper; the local module 32 is used to divide the ultra-high resolution portrait into a plurality of portrait blocks and then input them into the generator, each portrait block generates a local texture block under the guidance of the global albedo, and all the local texture blocks are assembled and mapped to the geometric body after post-processing to obtain a final texture map.

[0072] It should be noted that the texture map generation device based on physiological parameter decomposition provided in the above embodiment should be illustrated by the division of the above functional modules when performing texture map generation. The above functions can be assigned to different functional modules as needed, that is, the internal structure of the terminal or server is divided into different functional modules to complete all or part of the functions described above. In addition, the texture map generation device based on physiological parameter decomposition provided in the above embodiment and the texture map generation method embodiment based on physiological parameter decomposition belong to the same concept. The specific implementation process is detailed in the texture map generation method embodiment based on physiological parameter decomposition, which will not be repeated here.

[0073] Based on the same inventive concept, an embodiment further provides a computing device, including a memory and one or more processors, wherein executable code is stored in the memory, and when the one or more processors execute the executable code, the method for generating a texture map based on physiological parameter decomposition is implemented, specifically including the following steps:

[0074] S1, global stage: the ultra-high resolution portrait is downsampled and then passed through the texture mapping predictor to generate global texture mapping features. The global texture mapping features are position-encoded to obtain position encoding features and then passed through the physiological parameter predictor to generate multi-channel physiological parameter textures. The physiological parameter textures are mapped to global albedo by the mapper;

[0075] S2, local stage: The ultra-high-resolution portrait is divided into multiple portrait blocks and then input into the generator. Each portrait block generates a local texture block under the guidance of the global albedo, and all local texture blocks are assembled and mapped to the geometry to obtain the final texture map.

[0076] The computing device provided in the embodiment, in addition to the processor and memory, also includes hardware required for other services such as internal bus, network interface, memory, etc. at the hardware level. The memory is a non-volatile memory, and the processor reads the corresponding computer program from the non-volatile memory into the memory and then runs it to implement the texture map generation method based on physiological parameter decomposition described in S1-S2 above. Of course, in addition to the software implementation, the present invention does not exclude other implementation methods, such as logic devices or a combination of software and hardware, etc., that is to say, the execution subject of the following processing flow is not limited to each logic unit, but can also be hardware or logic devices.

[0077] Based on the same inventive concept, an embodiment further provides a computer-readable storage medium on which a program is stored. When the program is executed by a processor, the above-mentioned texture map generation method based on physiological parameter decomposition is implemented, which specifically includes the following steps:

[0078] S1, global stage: the ultra-high resolution portrait is downsampled and then passed through the texture mapping predictor to generate global texture mapping features. The global texture mapping features are position-encoded to obtain position encoding features and then passed through the physiological parameter predictor to generate multi-channel physiological parameter textures. The physiological parameter textures are mapped to global albedo by the mapper;

[0079] S2, local stage: The ultra-high-resolution portrait is divided into multiple portrait blocks and then input into the generator. Each portrait block generates a local texture block under the guidance of the global albedo, and all local texture blocks are assembled and mapped to the geometry to obtain the final texture map.

[0080] In the embodiment, computer-readable media include permanent and non-permanent, removable and non-removable media, and information storage can be implemented by any method or technology. The information can be computer-readable instructions, data structures, program modules or other data.

[0081] The specific implementation methods described above provide a detailed description of the technical solutions and beneficial effects of the present invention. It should be understood that the above is only the most preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, supplements and equivalent substitutions made within the scope of the principles of the present invention should be included in the protection scope of the present invention.

Claims

1. A texture map generation method based on physiological parameter decomposition, characterized in that: The following steps are involved: Global stage: After downsampling the ultra-high-resolution portrait, the global texture mapping feature is generated by the texture mapping predictor. The global texture mapping feature is position-encoded to obtain the position encoding feature, and then passed through the physiological parameter predictor to generate a multi-channel physiological parameter texture. The physiological parameter texture is mapped to the global albedo by the mapper; Local stage: The ultra-high-resolution portrait is divided into multiple portrait blocks and then input into the generator. Each portrait block generates a local texture block under the guidance of the global albedo, and all local texture blocks are assembled and mapped to the geometry to obtain the final texture map.

2. The method for generating texture maps based on physiological parameter decomposition according to claim 1, characterized in that: The physiological parameter predictor is constructed based on a neural network, and the multi-channel physiological parameter texture generated by the predictor includes melanin distribution, melanin type mixing ratio, and hemoglobin distribution.

3. The method for generating texture maps based on physiological parameter decomposition according to claim 1, characterized in that: When constructing a physiological parameter predictor based on a neural network, the loss function for parameter optimization includes content loss and smoothness loss, where the content loss is constructed based on the difference between the generated physiological parameter texture and the real physiological parameter texture, and the smoothness loss is constructed based on the horizontal and vertical gradients of the generated physiological parameter texture.

4. The method for generating texture maps based on physiological parameter decomposition according to claim 1, characterized in that: The mapper is constructed based on a neural network, which is used to map the global albedo of the skin based on the physiological parameter texture. The construction process is: First, according to formula (1) and formula (2), the lookup table of different physiological parameter value combinations to the epidermal absorption coefficient and the dermal absorption coefficient is calculated, and then the Monte Carlo random walk algorithm is used to calculate the albedo value under different absorption coefficients, thereby forming a mapping lookup table of different physiological parameter value combinations to albedo; (1)； (2)； in, is the light spectrum distribution, represents the absorption coefficient of eumelanin, represents the absorption coefficient of melanin, represents other absorption coefficients, represents the oxygenated hemoglobin absorption coefficient, represents the absorption coefficient of deoxyhemoglobin, represents the bilirubin absorption coefficient, Represents the distribution of melanin, Represents the ratio of melanin type mixture, represents the distribution of hemoglobin, represents the epidermal absorption coefficient, represents the dermis absorption coefficient, Represents the proportion of oxygenated hemoglobin in hemoglobin; Finally, the mapping lookup table is used as sample data to train the neural network to learn the mapping relationship between different physiological parameter value combinations and albedo to obtain the mapper.

5. The method for generating texture maps based on physiological parameter decomposition according to claim 1, characterized in that: The generator adopts a diffusion neural network. During the generation process, each portrait block is used as an initial vector, and in each time step of inverse diffusion, the global albedo of the corresponding position of the portrait block is used as a guiding condition and denoising noise is generated based on the neural network. De-noising is performed based on the denoising noise to obtain a local texture block.

6. The method for generating texture maps based on physiological parameter decomposition according to claim 1, characterized in that: The texture mapping predictor is constructed based on a neural network, the mapper is constructed based on a U-net network, and the physiological parameter predictor is constructed based on a diffusion neural network.

7. The method for generating a texture map based on physiological parameter decomposition according to any one of claims 1 to 6, characterized in that: Also includes: The generated multi-channel physiological parameter texture is edited to adjust the skin color, and the edited physiological parameter texture is mapped to the global albedo of different skin colors through a mapper.

8. A texture map generation device based on physiological parameter decomposition, characterized in that: include: A global module is used to downsample the ultra-high-resolution portrait and generate global texture mapping features through a texture mapping predictor. The global texture mapping features are position-encoded to obtain position encoding features and then passed through a physiological parameter predictor to generate multi-channel physiological parameter textures. The physiological parameter textures are mapped to global albedo through a mapper. The local module is used to divide the ultra-high-resolution portrait into multiple portrait blocks and input them into the generator. Each portrait block generates a local texture block under the guidance of the global albedo, and all local texture blocks are assembled and mapped to the geometry to obtain the final texture map.

9. A computing device comprising a memory and one or more processors, wherein the memory stores executable code, characterized in that: When the one or more processors execute the executable code, they are used to implement the texture map generation method based on physiological parameter decomposition according to any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that: A program is stored thereon, and when the program is executed by a processor, the texture map generation method based on physiological parameter decomposition described in any one of claims 1 to 7 is implemented.

Images