A Non-Photorealistic Approach to Night Image Enhancement Using Human JND

Night image enhancement is a challenging image enhancement problem due to its extremely low lighting condition and high dynamic range. Traditional methods such as gamma correction mostly try to keep the overall image lighting regime while increasing the exposure level. In this paper we propose a new non-photorealistic approach for this problem. The goal is to enhance each possible image detail regardless of the overall lighting condition, so as to see every possible image detail. In this approach, we first convert the image RGB color space to HSV color space and enhance only the V channel; Then we compute the background level for each pixel and locate all the edge points via human just-noticeable-difference (JND). Then we enhance all these edge points so that their background levels are adjusted to the range which is visually comfortable to humans. In the following step, we propagate these scaling factors to the rest of the non-edge points through least-squares optimization. We finally use the enhanced V channel and convert back to the RGB color space. Experimental results show that the propose approach, though not strictly photo-realistic, can achieve significant enhancement results entailing the most image details when compared with other typical methods.