Light field image restoration in low-light environment

Abstract

Light field (LF) imaging provides rich spatial and angular information, but is problematic in low-light environment as the images suffer from low contrast and visibility. In this paper, we present a learning-based method to enhance low-light LF images. A high-dimensional convolutional neural network (CNN) is introduced to extract the spatio-angular features from the LF. The network operates directly on the four-dimensional LF data rather than on individual sub-aperture images, avoiding the loss of geometric information. Color compensation is then performed on the enhanced LF images coming from the high-dimensional CNN to reduce color distortion. The experimental results show that the proposed method achieves noticeable improvement compared with state-of-the-art low-light image restoration techniques in both visual inspection and objective assessments.