Improvement of Low-Light Image by Convolutional Neural Network
Manbae Kim · 2019
Many researches have been carried out for enhancing low-light images over the past decades. One of the methods is Retinex theory, where reflectance component is recovered and illumination component is attenuated. Recently, hand-crafted approaches for low-light enhancement have been replaced by artificial neural networks. This paper presents a convolutional neural network that can replace the Retinx-based low-light enhancement method. Experiments carried out on 120 low-light images validated the feasibility of the replacement by producing satisfactory reflectance images.