Multi-scale dynamic fusion for correcting uneven illumination images

Images taken under non-ideal lighting conditions often suffer from uneven illumination, resulting in image distortion and unclear details. To address these issues, researchers have developed various methods for image enhancement. However, most of these methods are only applicable to specific types of images, especially those with underexposed exposures. Therefore, to achieve this goal, in this article, we propose a multi -scale dynamic fusion method for correcting uneven illumination images. This method can balance the overall illumination of unevenly illuminated images captured under non-ideal lighting conditions, while maintaining the original brightness contrast and enhancing the image details. We constructed a multi-branch multi-scale fully convolutional neural network, which uses attention mechanisms to focus on the bright and dark areas that need to be processed in the main branch, while the side branch is used to maintain the image's color and naturalness. Extracting features at different scales is beneficial for detail recovery, while a larger receptive field results in better brightness contrast and a more realistic visual experience. Finally, the attention fusion method is used to fuse the features of different branches to obtain the corrected results.

Automated summary

In this article, the authors propose a multi-scale dynamic fusion method for correcting images captured under non-ideal lighting conditions. The method balances the overall illumination, maintains brightness contrast, enhances image details, and provides a better visual experience through the use of a multi-branch multi-scale neural network, attention mechanisms, and fusion methods.