Perceptual Decolorization and Dehazing of Images and Videos

Abstract

The mastering craft of photography aspiration is to communicate the equivalent of what we saw and felt (Alfred Stieglitz). The key to obtain a satisfactory image representation lies in reproducing the expressive scene characteristics. Although photographic work is often regarded as a literal transcription of the reality, due to the depiction constrains, artists perform tedious work and employ many photographic controls to obtain a realistic representation. In this work we address the problem of enhancing images by contrast manipulation. The importance of contrast holds not just for the perceived dimensionality of lightness but for colors as well. Therefore, proper techniques for contrast manipulation in the image will automatically yield desired adjustments over perceived image appearance. The presented work proposes new ways of transforming the image color, style and appearances. Our work is motivated by several photographic and artistic techniques being validated by perceptual studies. In the first part, we introduce an algorithm that decolorize images and videos guided by the original saliency. The method is inspired by the Hering’s opponent process theory and aims to increase the contrast of the regions of interest, rather than over the entire image. We have based our approach on the assumption that preserving these salient regions in the converted image will result in a better preservation of the visual contrast and overall perceptual appearance. After the monochromatic luminance channel is filtered and stored as a reference, the luminance values are computed pixel-wise by mixing both saturation and hue values, creating a new spatial distribution with an increased contrast of the interest regions. All the pre-computed values are normalized in order to fit the entire intensity range while the intensity is re-balanced in order to conserve the local contrast in the initial image. Since our decolorization is accurate and preserves finest details, we can exploit variations in chromacity as well as luminance for application such as video decolorization, segmentation under different illuminants, detail enhancement, wide-baseline image matching and auditory substitution systems.