Variable fractional order-based structure-texture aware Retinex model with dynamic guidance illumination

Abstract

Image Retinex is developed for decomposition of an observed image into the illumination and reflectance components. In this paper, we introduce a general framework of variational model with dynamic guidance illumination for image Retinex, consisting of two coupled minimization problems. The first minimization problem is responsible for estimation of the illumination and reflectance components from the input image, and the other is used to dynamically update the guidance illumination under the control of the illumination prior. As a particular case of the proposed framework, we present an adaptive variable fractional order-based structure-texture aware Retinex model with dynamic guidance illumination. In the proposed model, the illumination prior is derived from the local maximum of the maximal RGB value in the input color image, followed by guided filtering. Qualitative and quantitative evaluations on three commonly-used datasets illustrate that the proposed model generally achieves higher performance in image decomposition with application to low-light enhancement, in comparison to several state-of-the-art Retinex-based models. In particular, ARISM and LOE metrics of the proposed model ranks in the top two across the three datasets.