An inverse isp image reconstruction method based on diffusion model

By using a diffusion model and a four-branch LoRA low-rank adaptation inverse ISP image reconstruction method, the problems of detail loss and cross-camera adaptation in existing technologies are solved, achieving high-quality RAW image reconstruction and improving structural consistency and generalization ability.

CN122289056APending Publication Date: 2026-06-26TIANJIN UNIV
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202610394467.1
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2026-03-28
Publication Date
2026-06-26

AI Technical Summary

Technical Problem

Existing inverse ISP methods are prone to detail loss, structural inconsistencies, or artifacts in high-frequency textures and edge regions. They also have weak cross-camera generalization capabilities and difficulty in simultaneously matching color and noise statistics, leading to color casts or inconsistent noise patterns.

Method used

An inverse ISP image reconstruction method based on a diffusion model is adopted. By constructing a diffusion model for RAW image reconstruction, the diffusion process is used for forward noise addition and reverse noise removal. Combined with four-branch LoRA low-rank adaptation, adaptive reconstruction of RAW domain differences from multiple cameras is achieved.

Benefits of technology

It improves training stability, reduces cross-camera distribution shift, enhances structural consistency, lowers the cost of multi-camera deployment, balances detail, color, and overall visual consistency, and generates high-quality RAW images.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN122289056A_ABST
    Figure CN122289056A_ABST
Patent Text Reader

Abstract

This invention discloses an inverse ISP image reconstruction method based on a diffusion model, belonging to the fields of computer vision and computational photography. This method uses pure noise as the initial input and introduces the target RGB image as a conditional constraint during the diffusion-based inverse denoising process. By fine-tuning the pre-trained diffusion model using ControlNet, the model learns the mapping relationship between RGB and RAW, thereby gradually guiding the reconstruction of a RAW domain image consistent with the conditional RGB content. To improve the structural consistency, detail fidelity, and color / brightness response of the reconstruction results, this invention incorporates attention-guided strategies (such as spatial attention and cross-feature interactive attention) during the conditional injection process. This fully utilizes the edge, texture, and semantic information in the RGB image, suppressing artifacts and detail loss caused by inversion / reconstruction, and enhancing the ability to restore details in extreme exposure areas such as highlights and shadows.
Need to check novelty before this filing date? Find Prior Art