一种联合局部和全局信息的图像重建系统及方法
By designing a mask autoencoder system that combines local and global information, and integrating edge detection and depth detection networks, the problem of ignoring detailed information in the image reconstruction process in existing technologies is solved, thereby improving the accuracy and efficiency of image reconstruction.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- NANJING UNIV OF POSTS & TELECOMM
- Filing Date
- 2023-11-24
- Publication Date
- 2026-07-17
AI Technical Summary
Existing mask autoencoder models tend to ignore detailed information such as image features and textures during image reconstruction, and the loss function only calculates the mean square error of image pixels, resulting in insufficient reconstruction accuracy and computational efficiency.
Design a masked autoencoder system that combines local and global information, including a generator network, an edge detection network, and a depth detection network. Constrain local and global information through edge detection loss functions and depth detection loss functions, and use a Transformer architecture and a deep convolutional network for image reconstruction.
It achieves the best balance between image reconstruction accuracy and computational efficiency, improving both the accuracy and efficiency of image reconstruction, and enabling better extraction of local details and global structure of images.
Smart Images
Figure CN117522674B_ABST