Image processing method, electronic device, and storage medium

CN118537243BActive Publication Date: 2026-11-03YUANLI TUXIN (CHONGQING) TECHNOLOGY CO LTD
View PDF 2 Cites 0 Cited by

Patent Information

Application Number
CN202410513431.1
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2024-04-26
Publication Date
2026-11-03
Estimated Expiration
2044-04-26

AI Technical Summary

Technical Problem

In existing technologies, deep learning shadow removal methods based on the Transformer architecture produce obvious ghosting in the generated images, while shadow removal methods based on the diffusion model output images with inconsistent illumination, resulting in poor shadow removal performance.

Method used

The shadow image, shadow mask image, and noisy image are segmented into slices of the target size. The shadow is removed by the global U-Net network in the diffusion model, and local processing is performed by combining it with the local U-Net network. The intermediate feature map of the global U-Net network is used to provide global guidance for the local U-Net network to generate the target image after shadow removal.

Benefits of technology

This improved shadow removal performance, ensured consistent illumination of the target image, prevented ghosting, and enhanced the shadow removal effect.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN118537243B_ABST
    Figure CN118537243B_ABST
Patent Text Reader

Abstract

Embodiments of the present application provide an image processing method, an electronic device and a storage medium. The method comprises: obtaining a shadow image, a shadow mask image and a noisy image to be processed; dividing the shadow image, the shadow mask image and the noisy image into target size slices respectively to obtain a plurality of groups of slices, each group of slices comprising a shadow slice, a shadow mask slice and a noisy slice corresponding in position; performing down-sampling processing on the shadow image and the shadow mask image respectively to obtain a global shadow image and a global shadow mask image of the target size; performing shadow removal processing on the global shadow image by a global U-Net network according to the global shadow mask image to obtain an intermediate feature map; processing each group of slices according to the intermediate feature map by a local U-Net network to obtain a local noise image; and generating a target image after shadow removal according to each group of local noise images and the noisy image. The embodiments of the present application improve the shadow removal effect.
Need to check novelty before this filing date? Find Prior Art

Citation Information

Patent Citations

  • Image shadow removal system and method based on potential diffusion model

    CN116485682A

  • Object class inpainting in digital images utilizing class-specific inpainting neural networks

    US20230368339A1