Three-dimensional face data acquisition device and three-dimensional face high-precision reconstruction method

Abstract

The application discloses a three-dimensional face data acquisition device, comprising: a transmitting device, an image acquisition device and a polarization adjusting device, the transmitting device is used for transmitting light to the collected object, the image acquisition device is used for collecting polarization data for three-dimensional reconstruction, and the polarization adjusting device is used for adjusting the polarization direction of polarized light; the transmitting device comprises a light source module, the light source module emits unpolarized white light for polarization, which is used by the polarization adjusting device; the image acquisition device comprises two color cameras, a first color camera is used for collecting cross-polarization data, and a second color camera is used for collecting parallel polarization data; the polarization adjusting device comprises a polarization direction adjusting device and an analyzing direction adjusting device; the polarization direction adjusting device is used together with the transmitting device to form a polarizer, and the white light is converted from unpolarized light to polarized light through the polarizer; the analyzing direction adjusting device is used together with the image acquisition device to form an analyzer, the direction of the analyzer is adjusted, and the polarized light corresponding to the direction is collected.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of three-dimensional face reconstruction, in particular to a low-cost and convenient three-dimensional face data acquisition device and a three-dimensional face high-precision reconstruction method using the acquisition device. BACKGROUND

[0002] With the rise of the concept of the metaverse, digital human model assets that improve the visual experience of metaverse users have gradually attracted people's attention. In particular, the establishment of digital human models for immersive social systems and virtual reality game systems and other applications not only meets the development needs of the metaverse, but also greatly improves the user experience of the metaverse.

[0003] Currently, the establishment of digital human assets with high precision often requires dozens of professional cameras combined with a certain physical framework to form a large camera array. The collector controls each camera in the array to take pictures at the same time, collects a large number of high-quality face images, and finally fits a high-precision three-dimensional face model according to the images. Professional operators use expensive and cumbersome acquisition equipment and go through a tedious and time-consuming process to successfully collect high-precision digital humans.

[0004] Therefore, for the existing digital human asset acquisition scheme, the camera array acquisition system requires a huge cost on the acquisition equipment; the cumbersome and difficult-to-move acquisition equipment limits the acquisition environment to be within the framework of the camera array. The dependence of the existing acquisition equipment on high-cost equipment and high-demand environments limits the promotion and popularization of such high-precision three-dimensional face model acquisition equipment and reconstruction schemes, further limiting the establishment of large-scale high-precision digital human assets and the development of the metaverse community. SUMMARY

[0005] The present application aims to overcome the shortcomings and deficiencies of the prior art and provide a three-dimensional face data acquisition device and a three-dimensional face high-precision reconstruction method. The three-dimensional face data acquisition device is low-cost and convenient, and the three-dimensional face high-precision reconstruction method can reconstruct a high-precision face model while meeting the requirements of low cost and convenience.

[0006] To achieve the above-mentioned purpose, the technical solutions adopted by the present application are as follows:

[0007] A three-dimensional face data acquisition device, comprising: a transmitting device, an image acquisition device, and a polarization adjustment device, the transmitting device is used to emit light to the object to be collected, the image acquisition device is used to collect polarization data for three-dimensional reconstruction, and the polarization adjustment device is used to adjust the polarization direction of polarized light;

[0008] The transmitting device includes at least two groups of light source modules, and the light source modules emit unpolarized white light for polarization for use by the polarization adjustment device;

[0009] The image acquisition device comprises at least two color cameras, a first color camera for acquiring cross-polarization data, and a second color camera for acquiring parallel-polarization data;

[0010] The polarization adjustment device comprises a polarization direction adjustment device and an analyzer direction adjustment device;

[0011] The polarization direction adjustment device is used together with the emission device to form a polarizer, and the white light is converted from non-polarized light to polarized light through the polarizer;

[0012] The analyzer direction adjustment device is used together with the image acquisition device to form an analyzer, and the polarized light of the corresponding direction is acquired by adjusting the direction of the analyzer.

[0013] A three-dimensional face high-precision reconstruction method applied to the three-dimensional face data acquisition device, comprising the following steps:

[0014] Acquiring expressionless and expressive polarization data and performing rough processing;

[0015] Reconstructing an expressionless face texture model based on expressionless polarization data;

[0016] Reconstructing an expressive face texture model based on expressive polarization data and the expressionless face texture model.

[0017] Preferably, acquiring expressionless and expressive polarization data and performing rough processing specifically comprises:

[0018] Acquiring data, using the three-dimensional face data acquisition device to acquire expressionless and expressive cross-polarization sequences, parallel-polarization sequences, cross-polarization pictures, and parallel-polarization pictures;

[0019] Roughly processing data, fitting the acquired data to form a three-dimensional face shape model, an initial diffuse reflection texture map, an initial specular reflection texture map, and an initial normal map.

[0020] Preferably, acquiring data specifically comprises the following steps:

[0021] Correctly installing the polarization adjustment device, so that the polarized white light emitted by the emission device is 90° different from the analyzer direction of the first color camera, and at the same time, the polarized white light emitted by the emission device is the same as the analyzer direction of the second color camera;

[0022] Adjusting the color and light amount of the first color camera and the second color camera;

[0023] Using the three-dimensional face data acquisition device, acquiring expressionless and expressive cross-polarization sequences, parallel-polarization sequences, cross-polarization pictures, and parallel-polarization pictures of the acquired object;

[0024] A high-precision three-dimensional face shape model, an initial diffuse reflection texture map, an initial specular reflection texture map, and an initial normal map are fitted using a model fitting software metashape by using expressionless and expression cross-polarization sequences, parallel polarization sequences, cross-polarization pictures, and parallel-polarization pictures.

[0025] Preferably, the expressionless face texture model is reconstructed, specifically comprising:

[0026] The skin is modeled using a spatially-varying bidirectional reflectance distribution function, and the skin reflection model is separated into a face diffuse reflection model and a face specular reflection model.

[0027] The face diffuse reflection model is generated, and a high-resolution diffuse reflection texture map and a normal map of the expressionless face skin are estimated using expressionless face cross-polarization data and the initial normal map.

[0028] The face specular reflection model is generated, and a high-resolution specular reflection texture and a final expressionless face normal map of the expressionless face skin are estimated using expressionless face parallel-polarization data, a high-resolution diffuse reflection texture and a normal map of the expressionless face skin.

[0029] Preferably, the face diffuse reflection model is generated, specifically comprising:

[0030] Features of the expressionless face cross-polarization data, the initial diffuse reflection texture map, and the initial normal map are extracted using a lightweight feature extraction network.

[0031] The features of the expressionless face cross-polarization data, the features of the initial normal map, and the expressionless face cross-polarization data are loaded into the face diffuse reflection model, and the initial diffuse reflection texture map and the initial normal map are optimized to obtain a high-resolution diffuse reflection texture map and a cross-polarization-optimized normal map of the expressionless face skin.

[0032] Preferably, the face specular reflection model is generated, specifically comprising:

[0033] Features of the expressionless parallel-polarization data, the initial specular reflection texture map, and the cross-polarization-optimized normal map are extracted using a lightweight feature extraction network.

[0034] The features of the expressionless face parallel-polarization data, the features of the cross-polarization-optimized normal map, and the expressionless face parallel-polarization data are loaded into the specular reflection model, and the initial specular reflection texture map and the cross-polarization-optimized normal map are optimized to obtain a high-resolution specular reflection texture map and a final expressionless face normal map of the expressionless face skin.

[0035] Preferably, the expression face texture model is reconstructed, specifically comprising:

[0036] A diffuse reflection model is established, and the high-resolution diffuse reflection texture difference map and displacement map of the skin of an expressive face are estimated using the cross-polarization data of the expressive face and the texture model of the neutral face.

[0037] A specular reflection model is established, and the high-resolution specular reflection difference map and final displacement map of the skin of the expressive face are estimated using the parallel polarization data of the expressive face, the texture model of the neutral face, the high-resolution diffuse texture difference map of the skin of the expressive face, and the displacement map.

[0038] The high-resolution diffuse reflectance texture difference map of the facial skin with expression is fused with the high-resolution diffuse reflectance texture map of the facial skin without expression to generate a high-resolution diffuse reflectance texture map of the facial skin with expression;

[0039] The high-resolution specular reflection texture difference map of the facial skin with expression is fused with the high-resolution specular reflection texture map of the facial skin without expression to generate a high-resolution specular reflection texture map of the facial skin with expression.

[0040] Preferably, establishing a diffuse reflection model specifically includes:

[0041] A lightweight feature extraction network is used to extract features of cross-polarization data of expressive faces and normal maps of neutral faces;

[0042] The features of the cross-polarization data of expressive faces, the features of the normal map of neutral faces, the cross-polarization data of neutral faces, and the cross-polarization data of expressive faces are loaded into the adversarial generative network to obtain a high-resolution diffuse texture difference map and displacement map of the skin of the expressive face.

[0043] Preferably, establishing a specular reflection model specifically includes:

[0044] A lightweight feature extraction network is used to extract features of parallel polarization data of expressive faces and normal maps of neutral faces;

[0045] The features of the parallel polarization data of the expressive face, the features of the normal map of the expressionless face, the parallel polarization data of the expressionless face, and the parallel polarization data of the expressive face are loaded into the adversarial generative network to obtain a high-resolution specular reflection texture difference map and a final displacement map of the skin of the expressive face.

[0046] Compared with the prior art, the present invention has the following advantages and beneficial effects:

[0047] 1. The three-dimensional face data acquisition device of the present invention can easily and separately acquire cross-polarization data and parallel polarization data, with low cost, convenience and speed.

[0048] 2. The three-dimensional face high-precision reconstruction method of the present application uses a three-dimensional face data acquisition device to acquire cross-polarization data and parallel polarization data of a face, decouples diffuse reflection and mirror reflection of the face, and obtains an optimized high-precision face texture model, so as to acquire data and reconstruct a high-precision face model under low cost and convenience requirements. BRIEF DESCRIPTION OF DRAWINGS

[0049] Figure 1 Fig. 1 is a structural schematic diagram of a three-dimensional face data acquisition device.

[0050] Figure 2 Fig. 5 is a flowchart of a three-dimensional face high-precision reconstruction method.

[0051] Figure 3 Fig. 6 is a flowchart of acquiring expressionless and expression polarization data and performing rough processing.

[0052] Figure 4 Fig. 7 is a flowchart of reconstructing an expressionless face texture model.

[0053] Figure 5 Fig. 8 is a framework schematic diagram of reconstructing an expressionless face texture model.

[0054] Figure 6 Fig. 9 is a flowchart of reconstructing an expression face texture model.

[0055] Figure 7 Fig. 10 is a framework schematic diagram of reconstructing an expression face texture model.

[0056] BRIEF DESCRIPTION OF DRAWINGS

[0057] 10 - emitting device; 101 - first light source module; 102 - second light source module; 20 - image acquisition device; 201 - first color camera; 202 - second color camera; 30 - polarization adjustment device; 301 - first polarization direction adjustment device; 302 - second polarization direction adjustment device; 303 - first detection direction adjustment device; 304 - second detection direction adjustment device. DETAILED DESCRIPTION

[0058] The three-dimensional face data acquisition device and the three-dimensional face high-precision reconstruction method of the present application will be further described below in combination with the drawings and specific embodiments.

[0059] Please refer to Figure 1 The present application discloses a three-dimensional face data acquisition device, which comprises an emitting device 10, an image acquisition device 20 and a polarization adjustment device 30. The emitting device 10 is used to emit light to a collected object. The image acquisition device 20 is used to acquire polarization data for three-dimensional reconstruction. The polarization adjustment device 30 is used to adjust the polarization direction of polarized light.

[0060] See also Figure 1 In this embodiment, the emitting device 10 includes a first light source module 101 and a second light source module 102. The first light source module 101 and the second light source module 102 emit unpolarized white light for polarization, which is used by the polarization adjustment device 30. The image acquisition device 20 includes a first color camera 201 and a second color camera 202. The first color camera 201 is used to acquire cross-polarization data, and the second color camera 202 is used to acquire parallel polarization data.

[0061] See also Figure 1 In this embodiment, the polarization adjustment device 30 includes a first polarization direction adjustment device 301, a second polarization direction adjustment device 302, a first polarization direction adjustment device 303, and a second polarization direction adjustment device 304. The first polarization direction adjustment device 301 is used together with the first light source module 101, and the second polarization direction adjustment device 302 is used together with the second light source module 102 to form a polarizer. White light passing through the polarizer is converted from unpolarized light to polarized light. The first polarization direction adjustment device 303 is used together with the first color camera 201, and the second polarization direction adjustment device 304 is used together with the second color camera 202 to form an analyzer. By adjusting the direction of the analyzer, polarized light in the corresponding direction is collected.

[0062] The three-dimensional face data acquisition device of the present invention can easily collect cross-polarization data and parallel polarization data separately, greatly reducing the data acquisition cost. These low-cost and conveniently collected data will provide data for reconstruction of the three-dimensional face model.

[0063] In this embodiment, the polarization adjustment device 30 is first correctly installed. The first polarization direction adjustment device 301 is installed on the first light source module 101, and the second polarization direction adjustment device 302 is installed on the second light source module 102. The first polarization direction adjustment device 303 is installed on the first color camera 201, and the second polarization direction adjustment device 304 is installed on the second color camera 202. The polarization directions of the first polarization direction adjustment device 301 and the first polarization direction adjustment device 303 are 90° apart from each other, forming a cross-polarization pair. The polarization directions of the second polarization direction adjustment device 302 and the second polarization direction adjustment device 304 are simultaneously aligned, forming a parallel polarization pair. The transmitting device 10 is activated to transmit polarized white light toward the object being captured, while the image capture device 20 is used to capture a color image. The cross-polarization pair formed by the first polarization direction adjustment device 301 and the first polarization direction adjustment device 303 collects cross-polarization data, while the parallel polarization pair formed by the second polarization direction adjustment device 302 and the second polarization direction adjustment device 304 collects parallel polarization data.

[0064] See also Figure 2The application further discloses a three-dimensional face high-precision reconstruction method.

[0065] S1: collecting expressionless and expression polarization data and performing rough processing;

[0066] S2: reconstructing an expressionless face texture model based on the expressionless polarization data;

[0067] S3: reconstructing an expression face texture model based on the expression polarization data and the expressionless face texture model.

[0068] The three-dimensional face high-precision reconstruction method of the application uses a three-dimensional face data acquisition device to collect cross-polarization data and parallel-polarization data of a face, decouples diffuse reflection and mirror reflection of the face, and obtains an optimized high-precision face texture model, so that a high-precision face model is reconstructed by collecting data under the requirements of low cost and convenience.

[0069] In step S1, the expressionless and expression polarization data are collected and rough processing is performed, and the specific steps include the following steps.

[0070] The data are collected by using a three-dimensional face data acquisition device to collect cross-polarization sequences, parallel-polarization sequences, cross-polarization pictures and parallel-polarization pictures of expressionless and expression faces.

[0071] The data are roughly processed to fit three-dimensional face shape models, initial diffuse reflection texture maps, initial mirror reflection texture maps and initial normal maps.

[0072] The expressionless refers to the expression of the eyes and mouth of a collector in a natural closed state, and the expression refers to the expression of the eyes and mouth of the collector in a non-natural state.

[0073] Please refer to Figure 3 The data are collected, and the specific steps include the following steps.

[0074] S11: correctly installing a polarization adjustment device, so that the polarization white light emitted by the emitting device is 90 degrees different from the detection direction of the first color camera, and at the same time, the polarization white light emitted by the emitting device is the same as the detection direction of the second color camera;

[0075] S12: adjusting the color and light amount of the first color camera and the second color camera;

[0076] S13: using a three-dimensional face data acquisition device to collect cross-polarization sequences, parallel-polarization sequences, cross-polarization pictures and parallel-polarization pictures of expressionless and expression faces of a collected object;

[0077] S14: Using the cross-polarization sequence with and without expression, the parallel polarization sequence, the cross-polarization picture and the parallel polarization picture, a high-precision three-dimensional face shape model, an initial diffuse reflection texture map, an initial specular reflection texture map and an initial normal map are fitted using model fitting software metashape.

[0078] Please refer to Figure 4 In step S2, the expressionless face texture model is reconstructed, specifically including the following steps:

[0079] S21: The skin is modeled using a spatially varying bidirectional reflectance distribution function (SV-BRDF), and the skin reflection model is separated into a face diffuse reflection model and a face specular reflection model;

[0080] S22: The face diffuse reflection model is generated, and the high-resolution diffuse reflection texture map and the normal map of the expressionless face skin are estimated using the expressionless face cross-polarization data and the initial normal map;

[0081] S23: The face specular reflection model is generated, and the high-resolution specular reflection texture and the final expressionless face normal map of the expressionless face skin are estimated using the expressionless face parallel polarization data, the high-resolution diffuse reflection texture and the normal map of the expressionless face skin.

[0082] Please refer to Figure 5 In step S22, when the face diffuse reflection model is generated, the features of the expressionless face cross-polarization data, the features of the initial diffuse reflection texture map and the features of the initial normal map are extracted using a lightweight feature extraction network; the extracted features of the expressionless face cross-polarization data, the features of the initial normal map and the expressionless face cross-polarization data are loaded into the diffuse reflection model, and the initial diffuse reflection texture map and the initial normal map are optimized to obtain the high-resolution diffuse reflection texture map of the expressionless face skin and the cross-polarization optimized normal map.

[0083] Please refer to Figure 5 In step S23, when the face specular reflection model is generated, the features of the expressionless parallel polarization data, the features of the initial specular reflection texture map and the features of the cross-polarization optimized normal map are extracted using a lightweight feature extraction network; the extracted features of the expressionless face parallel polarization data, the extracted features of the cross-polarization optimized normal map and the expressionless face parallel polarization data are loaded into the specular reflection model, and the initial specular reflection texture map and the cross-polarization optimized normal map are optimized to obtain the high-resolution specular reflection texture map of the expressionless face skin and the final expressionless face normal map.

[0084] Please refer toFigure 6 In step S3, the textured model of the facial expression is reconstructed, and the steps include the following steps.

[0085] S31: A diffuse reflection model is established, and the high-resolution diffuse reflection texture difference map and the displacement map of the facial expression skin are estimated by using the cross-polarization data of the facial expression and the textured model of the facial expressionless.

[0086] S32: A specular reflection model is established, and the high-resolution specular reflection difference map and the final displacement map of the facial expression skin are estimated by using the parallel-polarization data of the facial expression, the textured model of the facial expressionless, the high-resolution diffuse reflection texture difference map and the displacement map of the facial expression skin.

[0087] S33: The high-resolution diffuse reflection texture difference map of the facial expression skin is fused with the high-resolution diffuse reflection texture map of the facial expressionless skin to generate the high-resolution diffuse reflection texture map of the facial expression skin.

[0088] The high-resolution specular reflection texture difference map of the facial expression skin is fused with the high-resolution specular reflection texture map of the facial expressionless skin to generate the high-resolution specular reflection texture map of the facial expression skin.

[0089] Please refer to Figure 7 In step S31, when the diffuse reflection model is established, the features of the cross-polarization data of the facial expression and the features of the normal map of the facial expressionless are extracted by using the lightweight feature extraction network; the extracted features of the cross-polarization data of the facial expression, the extracted features of the normal map of the facial expressionless, the cross-polarization data of the facial expressionless and the cross-polarization data of the facial expression are loaded into the generative adversarial network to obtain the high-resolution diffuse reflection texture difference map and the displacement map of the facial expression skin.

[0090] Please refer to Figure 7 In step S32, when the specular reflection model is established, the features of the parallel-polarization data of the facial expression and the features of the normal map of the facial expressionless are extracted by using the lightweight feature extraction network; the extracted features of the parallel-polarization data of the facial expression, the extracted features of the normal map of the facial expressionless, the parallel-polarization data of the facial expressionless and the parallel-polarization data of the facial expression are loaded into the generative adversarial network to obtain the high-resolution specular reflection texture difference map and the final displacement map of the facial expression skin.

[0091] In summary, the present application has the following advantages and beneficial effects:

[0092] 1. The three-dimensional facial data acquisition device of the present application can simply and conveniently acquire cross-polarization data and parallel-polarization data, and has low cost and is convenient and fast.

[0093] 2. The three-dimensional face high-precision reconstruction method of the present application uses a three-dimensional face data acquisition device to acquire cross-polarization data and parallel-polarization data, decouples the human face diffuse reflection and the human face mirror reflection, and obtains an optimized high-precision human face texture model, so as to acquire data and reconstruct a high-precision human face model under low cost and convenience requirements.

[0094] The above description is a detailed description of the preferred embodiments of the present application, but the embodiments are not intended to limit the scope of the patent application of the present application. Any equivalent changes or modifications made under the technical spirit disclosed by the present application should belong to the patent scope covered by the present application.

Claims

1. A method for high-precision reconstruction of three-dimensional face, applied to a three-dimensional face data acquisition device, and characterized in that: the three-dimensional face data acquisition device comprises a transmitting device, an image acquisition device and a polarization adjustment device, the transmitting device is used to emit light to a collected object, the image acquisition device is used to collect polarization data for three-dimensional reconstruction, and the polarization adjustment device is used to adjust the polarization direction of polarized light; the transmitting device comprises at least two groups of light source modules, the light source modules emit unpolarized white light for polarization, and the polarization adjustment device uses the unpolarized white light; the image acquisition device comprises at least two color cameras, a first color camera is used to collect cross-polarization data, and a second color camera is used to collect parallel-polarization data; the polarization adjustment device comprises a polarization direction adjustment device and an analyzer direction adjustment device; the polarization direction adjustment device is used together with the transmitting device to form a polarizer, and white light is converted from unpolarized light to polarized light through the polarizer; the analyzer direction adjustment device is used together with the image acquisition device to form an analyzer, and polarization light of a corresponding direction is collected by adjusting the direction of the analyzer; the method for high-precision reconstruction of three-dimensional face comprises the following steps: collecting expressionless and expressive polarization data and performing rough processing; reconstructing an expressionless face texture model based on expressionless polarization data; reconstructing an expressive face texture model based on expressive polarization data and the expressionless face texture model; and the reconstruction of the expressive face texture model specifically comprises: establishing a diffuse reflection model, estimating a high-resolution diffuse reflection texture difference map and a displacement map of the expressive face skin by using the cross-polarization data of the expressive face and the expressionless face texture model; establishing a specular reflection model, estimating a high-resolution specular reflection difference map and a final expressive displacement map of the expressive face skin by using the parallel-polarization data of the expressive face, the expressionless face texture model, the high-resolution diffuse reflection texture difference map and the displacement map of the expressive face skin; fusing the high-resolution diffuse reflection texture difference map of the expressive face skin with a high-resolution diffuse reflection texture map of the expressionless face skin to generate a high-resolution diffuse reflection texture map of the expressive face skin; and fusing the high-resolution specular reflection texture difference map of the expressive face skin with a high-resolution specular reflection texture map of the expressionless face skin to generate a high-resolution specular reflection texture map of the expressive face skin. The collection of expressionless and expressive polarization data and the rough processing specifically comprise: collecting data, collecting cross-polarization sequences, parallel-polarization sequences, cross-polarization pictures and parallel-polarization pictures of expressionless and expressive faces by using the three-dimensional face data acquisition device; and rough processing data, fitting a three-dimensional face shape model, an initial diffuse reflection texture map, an initial specular reflection texture map and an initial normal map from the collected data. The collection of data specifically comprises the following steps: correctly installing the polarization adjustment device, making the polarized white light emitted by the transmitting device differ by 90° from the analyzer direction of the first color camera, and at the same time making the polarized white light emitted by the transmitting device the same as the analyzer direction of the second color camera; adjusting the color and light amount of the first color camera and the second color camera; and adjusting the color and light amount of the first color camera and the second color camera. ​ ​ ​ ​ ​ ​ ​ ​ ​ ​ ​ ​ ​ ​ ​ 2. The high-precision 3D face reconstruction method according to claim 1, characterized in that: ​ ​ ​ 3. The method of claim 2, wherein, ​ ​ ​ The three-dimensional face data acquisition device is used to collect the expressionless and expression cross-polarization sequence, parallel polarization sequence, cross-polarization picture and parallel polarization picture of the collected object; The expressionless and expression cross-polarization sequence, parallel polarization sequence, cross-polarization picture and parallel polarization picture are used to fit a high-precision three-dimensional face shape model, an initial diffuse reflection texture map, an initial specular reflection texture map and an initial normal map by using a model fitting software metashape.

4. The method of claim 1, wherein, The expressionless face texture model is reconstructed, specifically comprising: The skin is modeled by using a spatially varying bidirectional reflectance distribution function, and the skin reflection model is separated into a face diffuse reflection model and a face specular reflection model; The face diffuse reflection model is generated, and the high-resolution diffuse reflection texture map and the normal map of the expressionless face skin are estimated by using the expressionless face cross-polarization data and the initial normal map; The face specular reflection model is generated, and the high-resolution specular reflection texture and the final expressionless face normal map of the expressionless face skin are estimated by using the expressionless face parallel polarization data, the high-resolution diffuse reflection texture and the normal map of the expressionless face skin.

5. The method of claim 4, wherein, The face diffuse reflection model is generated, specifically comprising: The features of the expressionless face cross-polarization data, the initial diffuse reflection texture map and the initial normal map are extracted by using a lightweight feature extraction network; The features of the expressionless face cross-polarization data, the initial normal map and the expressionless face cross-polarization data are loaded into the face diffuse reflection model, and the initial diffuse reflection texture map and the initial normal map are optimized to obtain the high-resolution diffuse reflection texture map of the expressionless face skin and the cross-polarization optimized normal map.

6. The method of claim 4, wherein, The face specular reflection model is generated, specifically comprising: The features of the expressionless parallel polarization data, the initial specular reflection texture map and the cross-polarization optimized normal map are extracted by using a lightweight feature extraction network; The features of the expressionless face parallel polarization data, the cross-polarization optimized normal map and the expressionless face parallel polarization data are loaded into the specular reflection model, and the initial specular reflection texture map and the cross-polarization optimized normal map are optimized to obtain the high-resolution specular reflection texture map of the expressionless face skin and the final expressionless face normal map.

7. The method of claim 1, wherein, The diffuse reflection model is established, specifically comprising: The features of the expressionless face cross-polarization data and the expressionless face normal map are extracted by using a lightweight feature extraction network; The features of the expressionless face cross-polarization data, the expressionless face normal map, the expressionless face cross-polarization data and the expression face cross-polarization data are loaded into the generative adversarial network to obtain the high-resolution diffuse reflection texture difference map and the displacement map of the expression face skin.

8. The method of claim 1, wherein, The specular reflection model is established, specifically comprising: The features of the expressionless face parallel polarization data and the expressionless face normal map are extracted by using a lightweight feature extraction network; The features of the expressionless face parallel polarization data, the expressionless face normal map, the expressionless face parallel polarization data and the expression face parallel polarization data are loaded into the generative adversarial network to obtain the high-resolution specular reflection texture difference map and the final displacement map of the expression face skin.

Images