Improved Single Image Haze Removal by Multi Scale Fusion

Abstract

The visibility of outdoor images can be degraded by the hazy weather conditions. A modified multi-scale fusion method is proposed in this paper to remove dense homogenous haze as well as fog present in an image. It is a fusion based strategy by deriving two hazy image inputs which is obtained by applying white balancing and a contrast enhancing procedure. Then filter their important features by applying weight maps which are luminance, chromaticity and saliency to preserve the important features of the derived inputs. These weight maps introduce artifacts in the images. To eliminate these artifacts we apply a multi scale fusion technique which is obtained by Gaussian and Laplacian pyramids. Then to eliminate the lightness of the image, we apply an image enhancement method. It is based on retinex theory and lightness decomposition method. The proposed method is then compared with the existing multi-scale fusion method. The results obtained shows that the new method enhances the image visibility in dense hazy regions.