Contrast enhancement in poor visibility conditions

Abstract

Today’s computer vision system is designed to operate on a clear day. Needless to say, in any outdoor scene, there is no escape from poor visibility conditions. Poor visibility conditions that arise in adverse weather impede the performance of vision systems because of the contrast loss, reduced visibility and color distortion. These characteristics make the contrast enhancement a very difficult job since the degradation is spatially-variant. Various model and non-model based contrast enhancement methods exist in literature to enable image applications to work reliably in poor visibility. Model based methods improve image contrast by reversing the underlying cause of image degradation whereas non-model based algorithms require no information about the cause of deterioration. This research work proposes two methods to restore the contrast in poor visibility. First is a model based method that enhances the contrast of weather-degraded images by extracting the atmospheric veil. The proposed algorithm based on guided filtering can accurately recover hidden edges, maintains structural similarity to input image and avoid oversaturation in colors. Experimental comparisons with state-of-the-art algorithms demonstrate that proposed approach can significantly enhance the contrast and retrieve the visibility in fine details without generating haloing effect. Second proposed method is non-model based, prompted by multiscale retinex with color restoration technique that allows vision applications to perform robustly in poor lighting conditions.