An Effective Low-Light Image Enhancement Algorithm via Fusion Model

Abstract

In a low-light condition, the quality of a captured image may be much poorer than that obtained in a normal environment. As an effective preprocessing step, many enhancement algorithms have been proposed to improve the performance of a computer vision task. In most existing algorithms, a image is often enhanced as a whole. As a result, the image may be over-enhanced or under-enhanced due to different degree of exposure in local area. Aiming at this issue, in this paper, we propose a low-light image enhancement algorithm based on image fusion technology. In the proposed method, a fusion strategy is devised by considering the exposure extent of local area. The weight matrix for image fusion is first calculated. Then, the pixel with insufficient exposure is selected according to the adaptive threshold. Next, the multi-exposure images can be synthesized by using the estimated optimal exposure rate. Finally, we use the input image to fuse with the enhanced image for slightly under-exposed images to get the enhancement image, while the severely under-exposed images can be enhanced by fusing a reflection map based on retinex with the enhanced image. Experimental results show that our method can obtain enhancement results with less color and lightness distortion compared to several state-of-the-art methods.