基于区域掩码与动态退出的扩散模型图像复原方法
By adopting a diffusion model-based image restoration method with region masking and dynamic exit, the problems of high computational overhead and iterative redundancy in existing technologies are solved, achieving efficient and accurate image restoration results.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- TAIYUAN UNIVERSITY OF TECHNOLOGY
- Filing Date
- 2025-12-19
- Publication Date
- 2026-07-17
AI Technical Summary
Existing diffusion models suffer from high computational overhead and redundant iteration steps in image restoration tasks. This is especially true when only certain regions require fine denoising and restoration, resulting in significant computational waste for the entire image. Furthermore, a fixed number of iteration steps cannot optimize the restoration quality.
An image restoration method based on a diffusion model with region masking and dynamic termination is adopted. Latent spatial masks are generated through edge detection. UNet denoising and ControlNet conditional constraints are applied only to the effective regions, and the iteration is dynamically terminated to meet the requirements of quality and consistency.
The computational load of the effective region is reduced by 30%-60%, the inference speed is increased by 2-3 times, the memory usage is reduced, the structure restoration accuracy of the effective region is improved by 15%-25%, the consistency is improved by 20%, and the computational interference and iteration redundancy of the invalid region are avoided.
Smart Images
Figure CN121860872B_ABST