Abstract

This paper describes a color restoration technique used to remove NIR information from single sensor cameras where color and near-infrared images are simultaneously acquired—referred to in the literature as RGBN images. The proposed approach is based on a neural network architecture that learns the NIR information contained in the RGBN images. The proposed approach is valuated on real images obtained by using a pair of RGBN cameras. Additionally, qualitative comparisons with a naive color correction technique based on mean square error minimization are provided.