Images taken in low-light environments are vulnerable to low visibility, which can also reduce the efficiency of several applications for computer vision and computational photography.Images can be easily captured by different image acquisition devices these days. Low-illumination images will be produced by weak lighting conditions and technologies with poor filling flash.It is difficult to classify these damaged images, and certain approaches should be handled via the computer.A new semantic image segmentation based on the deep learning techniques was proposed for improving the visibility of images captured in the low illumination environment.An improved deep learning approach to segment low-illumination images is proposed in this paper, based on existing CNN research on the low illumination environment. On low illumination images with mixed noises, the robustness and efficiency of the proposed system are evaluated.Results show that other techniques of image segmentation are outperformed by the proposed technique.

Image Segmentation; Deep Learning; Low illumination