Contrast enhancement based on reflectance-oriented probabilistic equalization

Abstract

Despite advances in contrast enhancement, it remains difficult to adaptively improve both the global contrast (base structures) and the local contrast (detailed textures) of an image. To solve this problem, we first propose a novel 2D histogram equalization approach called ROPE for global contrast enhancement. ROPE assumes intensity occurrence and co-occurrence to be dependent on each other and newly estimates the distribution of intensity occurrence (1D histogram) by marginalizing over the distribution of intensity co-occurrence (2D histogram). The 2D histogram is generated by incorporating local reflectance differences into the density estimation, so that ROPE can provide sufficient brightness enhancement not only for normal-light images but also for dark images. For local contrast enhancement, ROPE is extended to compute locally adaptive 2D histograms, giving birth to a novel approach called AdaROPE. Extensive experiments on four public datasets consisting of 8400 images demonstrate the superiority of our approaches over state-of-the-art image enhancement.