Deep Retinex Network for Estimating Illumination Colors with Self-Supervised Learning

Abstract

We propose a novel Retinex image-decomposition network that can be trained in a self-supervised manner. The Retinex image-decomposition aims to decompose an image into illumination-invariant and illumination-variant components, referred to as "reflectance" and "shading," respectively. Although there are three consistencies that the reflectance and shading should satisfy, most conventional work considers only one or two of the consistencies. For this reason, the three consistencies are considered in the proposed network. In addition, by using generated pseudo-images for training, the proposed network can be trained with self-supervised learning. Experimental results show that our network can decompose images into reflectance and shading components. Furthermore, it is shown that the proposed network can be used for white-balance adjustment.

Discriminative feature encoding for intrinsic image decomposition

“…Facing the fact that high-quality and denselylabelled intrinsic images are expensive to acquire, many methods have been developed to train models with additional constraints [16,42], reusing physically motivated priors [1,[43][44][45][46], expanding the dataset with synthetic images [15,17,47], or training across datasets [17].…”

“…By utilizing additional constraints, the solution space for the problem is further restricted. Seo et al [43] proposed an imagedecomposition network, which makes use of all the three premises regarding consistency from retinex theory. In this work, pseudo images (generated color-transferred multi-exposure images) are used for training.…”

Separated-Spectral-Distribution Estimation Based on Bayesian Inference with Single RGB Camera

“…Reflectance r j and sensitivity c k can be estimated in the same way as for illumination [see eqs. (11) to (16)].…”

“…Spectral distribution is a function of wavelengths and describes what wavelengths of light are included in illumination, reflected by objects, or detected by imaging sensors. Knowledge on spectral distributions is essential for many computer vision tasks that use color information, such as hyperspectral imaging [1,2], color constancy [3][4][5][6][7][8], color image enhancement [9][10][11][12], and intrinsic image decomposition [13][14][15][16].…”