An Efficient Illumination Estimation Algorithm for Low-Light Image Enhancement

Low-light image enhancement is of interest in many practical applications, such as low-light photography and video surveillance. Traditional methods, like histogram equalization-based methods, cannot produce good quality output consistently. Recently, a Retinex-based method called LIME has been proposed which can produce good quality output robustly without any learning procedures. The key step of this algorithm is illumination map refinement, which is achieved by solving an optimization problem. Inspired by LIME, we propose a simple and fast way to carry out the illumination map refinement by using guided image filtering and superpixel segmentation without any optimization processes, which can achieve comparable enhancement performance to LIME while running at several times the speed of the fast approximation solution of LIME. Furthermore, we integrate an enhancement control parameter into the refinement process to mitigate the over-enhancement problem. Experimental results show that the proposed method can not only retain the original input color distribution very well, but also obtain comparable image quality scores to state-of-the-art neural network-based algorithms.