Method, apparatus and electronic device for generating virtual image
By integrating local and global facial feature information into a 3D face generation model and combining it with a self-attention mechanism to process images, the problem of low realism of virtual images in the metaverse is solved, achieving realistic 3D face model generation and improving user experience.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- INDUSTRIAL AND COMMERCIAL BANK OF CHINA
- Filing Date
- 2023-04-28
- Publication Date
- 2026-07-21
AI Technical Summary
Existing technologies generate virtual images of customers in the metaverse with low realism, and traditional face reconstruction processes do not fully consider lighting and facial expression factors, resulting in poor quality of reconstructed 3D face models.
A 3D face generation model is adopted. By fusing local and global face feature information and combining a self-attention mechanism, illumination feature information is extracted from the target image. Multiple convolutional and pooling layers are used to process the image, generate a 3D face feature vector, and generate a virtual image in the metaverse.
It improves the realism of the virtual avatar of the customer in the metaverse by extracting and fusing local and global facial feature information to generate a realistic 3D facial model, thereby enhancing the user experience.
Smart Images

Figure CN116524564B_ABST