Due to the noise in the input image, the ability of image semantic segmentation may deteriorate. In this case, enhancing the image before segmentation can be helpful. This article found that not only does image enhancement help solve the problem of segmentation accuracy decline caused by noise, but also pixelated semantic information can improve image enhancement ability. Therefore, this article proposes a fast single-image processing method that combines image semantic segmentation with image enhancement. A segmentation-enhancement alternating boosting network is designed. The network consists of multiple segmentation and enhancement modules. The synergistic effect between image enhancement and semantic segmentation is explored through this alternating boosting network. It was found through experiments that as the number of segmentation-enhancement modules increased, both enhancement and segmentation performance improved. The quality of the enhanced image was greatly improved, and the segmentation accuracy was also improved to be close to that of a clean image. This indicates that segmentation improves image enhancement performance, while image enhancement helps improve segmentation accuracy.
Algorithm for Single Image Enhancement Based on Semantic Segmentation Assistance
2023-10-11
2478063 byte
Conference paper
Electronic Resource
English