Raw domain image denoising method and device based on artificial intelligence, equipment and medium

By using physical modeling of sensor imaging noise and an artificial intelligence model based on the U-Net architecture, the motion blur problem caused by multi-frame synthesis in existing technologies is solved, achieving efficient single-frame image denoising and improving the accuracy and reliability of image denoising.

CN122265084APending Publication Date: 2026-06-23MALANSHAN AUDIO & VIDEO LABORATORY
View PDF 0 Cites 0 Cited by

Patent Information

Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
MALANSHAN AUDIO & VIDEO LABORATORY
Filing Date
2026-04-23
Publication Date
2026-06-23

AI Technical Summary

Technical Problem

Existing image denoising methods suffer from motion blur due to multi-frame synthesis, making it difficult to achieve efficient single-frame denoising.

Method used

By using physical modeling of sensor imaging noise to generate prior information, an artificial intelligence model based on the U-Net architecture is constructed. The model is then optimized using a training set to achieve noise removal from single-frame images.

Benefits of technology

It achieves single-frame denoising, avoids motion blur, improves the accuracy and reliability of image denoising, and reduces the computational load and deployment difficulty of the model.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN122265084A_ABST
    Figure CN122265084A_ABST
Patent Text Reader

Abstract

The application discloses an artificial intelligence-based RAW domain image denoising method and device, equipment and medium, relates to the technical field of image processing, and comprises the following steps: generating a target training set by using target prior information; training a target denoising model by using the target training set to obtain a trained denoising model; obtaining an original RAW domain noise image output by a preset sensor, extracting image noise corresponding to the original RAW domain noise image by using the trained denoising model, and subtracting the original RAW domain noise image from the image noise to realize denoising of the original RAW domain noise image. By using the model to obtain the image noise of a single frame of image and subtracting the original image from the noise, single-frame denoising is realized, and trailing is avoided. By generating training data according to the prior information of sensor imaging noise and training the model, the model does not need to learn the noise distribution from zero, and the problem of difficult lightweight deployment of the model is solved.
Need to check novelty before this filing date? Find Prior Art