一种图像上色模型构建方法、图像上色方法、设备及介质
By constructing a grid latent feature and binary mask combined with a color prior prediction module and an alignment loss function, the problems of inaccurate region localization, missing semantic coverage, and color misalignment in text-guided image coloring are solved, and high-fidelity, semantically consistent color image generation is achieved.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- NANJING UNIV OF POSTS & TELECOMM
- Filing Date
- 2026-04-09
- Publication Date
- 2026-07-17
AI Technical Summary
Existing technologies for text-guided image coloring suffer from problems such as inaccurate region positioning, missing semantic coverage, and color overflow and misalignment, making it difficult to generate high-fidelity and semantically consistent color images.
An image colorization method based on a diffusion model is adopted. By constructing grid latent features and binary masks, combined with a color prior prediction module and a color-object alignment loss function, it achieves accurate pixel-level localization and complete semantic coverage, and eliminates color misalignment interference.
It significantly improves the color fidelity and semantic consistency of generated color images, reduces training costs, achieves color completion and enrichment from the full-view perspective, and enhances the practicality of the model.
Smart Images
Figure CN121999090B_ABST