A Comparative Study on CNN based Low-light Image Enhancement

Images collected under low-light conditions have an intrinsic property of poor illumination, due to the environment that they're captured in. Improving the quality of images pro-poses various challenges such as low contrast, low brightness, noise etc. Simply increasing the brightness of the dark regions will inevitably amplify the hidden artifacts. Deep Learning has shown great promise in this task, and it's black-box nature is used heavily for the enhancement of under-exposed images. In this paper we present a comparative analysis of various techniques based on Convolutional Neural Networks (CNN) that aim to enhance under-exposed images and perform extensive experiments on these techniques. The performance of these algorithms are analyzed subjectively and quantitatively using established metrics.