3D Face Model Texture Reconstruction Using Multi-View Fusion
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing image processing methods for face reconstruction in 3D virtual reality, augmented reality, and game applications produce texture images with low accuracy due to rough texture image reconstruction.
Innovation Solution
The method involves projecting a face model onto multiple face images from different view angles to determine correspondences between regions, extracting relevant images based on these correspondences, and fusing them to generate a more accurate texture image using algorithms like Laplacian pyramid fusion.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Manufacturing precision
If a texture image reconstruction solution is used in related art, then the face reconstruction can be achieved, but the texture image accuracy is low and the reconstruction effect is rough
Solution Approach 1:
The patent segments the face image into multiple regions (forehead, eyes, nose, mouth, chin, etc.) and processes each region separately using different extraction strategies. This segmentation allows for more precise control over texture extraction in different facial areas, improving overall texture image accuracy while maintaining reconstruction quality.
Solution Approach 2:
The patent transforms the 2D face images into a 3D texture space by establishing correspondence relationships between 2D image coordinates and 3D model vertices. This dimensional transformation enables accurate mapping of texture information from multiple view angles onto the 3D face model, significantly improving texture accuracy and reconstruction effect.
2Manufacturing precision
If multiple face images from different view angles are processed, then the texture image accuracy can be improved, but the computational complexity and processing time increase
Solution Approach 1:
The patent performs preliminary actions by pre-establishing the correspondence relationship between 2D face images and the 3D face model before actual texture extraction. View angle parameters are pre-calculated and stored, allowing for efficient texture extraction from multiple views without repeated complex computations, thus reducing processing complexity while maintaining high accuracy.
Solution Approach 2:
The patent uses a standardized 3D face model as a template or copy that serves as the target for texture mapping. This template model with predefined vertex structure and region segmentation allows consistent processing of multiple input images, simplifying the overall processing pipeline while achieving accurate texture reconstruction from various view angles.
Data Source
AI summary
An image processing method and apparatus, a computer-readable medium, and an electronic device are provided. The image processing method includes: respectively projecting, according to a plurality of view angle parameters corresponding to a plurality of view angles, a face model of a target object onto a plurality of face images of the target object acquired from the plurality of view angles, to determine correspondences between regions on the face model and regions on the face image; respectively extracting, based on the correspondences and a target region in the face model that need to generate a texture image, images corresponding to the target region from the plurality of face images; and fusing the images that correspond to the target region and that are respectively extracted from the plurality of face images, to generate the texture image.