Enhancement of Low Light Images via IIIumination Map Estimation using Variational Optimization-based Retinex Model

Retinex is a method of bridging the gap between images and the human observation of scenes. Retinex theory is a model of lightness and color perception of human vision proposed by Edwin Land in 1986. While the retinex theory was actually aimed at providing explanation of human color perception, it has led to various image enhancement algorithms called as retinex algorithms, which are usually used to enhance local image contrast. Single-scale Retinex (SSR) is defined as an implementation of center/surround Retinex. Superposition of weighted different scale SSR balances both dynamic range compression and tonal rendition, which is Multiscale Retinex (MSR). For color images, spatial averages of the three color bands are far from equal, thus the output appears grey. To address this issue, a weight factor for different channels is introduced which is Multiscale Retinex with color restoration (MSRCR). In this paper, SSR, MSR and MSRCR systems for image enhancement are implemented and their performances are compared using MATLAB as the software tool.