图像处理方法、电子设备、存储介质及计算机程序产品
By processing the original image using a pre-trained illumination enhancement model, the image enhancement problem of the traditional Retinex model is solved, achieving an image enhancement effect that is closer to real lighting.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- YUANLI TUXIN (CHONGQING) TECHNOLOGY CO LTD
- Filing Date
- 2022-10-11
- Publication Date
- 2026-07-17
AI Technical Summary
Traditional image illumination enhancement methods based on the Retinex model suffer from problems such as impractical reflection component assumptions, noise amplification, artifacts and color deviations, and uneven illumination images, resulting in enhanced images that cannot closely resemble real images.
A pre-trained illumination enhancement model is used to process the original image and the initial illumination image. The model uses the sample standard image as the learning target and performs illumination enhancement based on the sample initial illumination image. This includes the first sub-model decomposition, the second sub-model adjustment, the enhancement module enhancement, and the third sub-model fusion to form the illumination-enhanced image.
The resulting image with enhanced illumination closely resembles the real image under standard illumination, avoiding the shortcomings of the Retinex model and achieving image enhancement that more closely resembles real lighting conditions.
Smart Images
Figure CN115841427B_ABST