STEP UP IN IMAGE DEHAZING USING WAVELET DECOMPOSITION & GUIDED FILTERS

Abstract

In the present work, haze removal of images based on dark channel prior approach has been proposed. Based on the observation that haze impacts low frequency component of the image, only approximate part of the hazy image extracted using lifting DB4 wavelet is processed by proposed dark channel prior algorithm. Haze imaging model along with dark channel prior has been used to estimate atmospheric light and transmission. Patch size used for determining dark channel has been made adaptive on the basis of image size. Besides, airlight is obtained from larger patch to ensure more accurate estimation in the presence of localized light sources. Further, transmission has been estimated separately for each of the three color channels considering the phenomena that different color channels undergo different scattering in the atmosphere depending upon their wavelengths. In addition to this, transmission of near white objects has been selectively increased to prevent them from getting over saturated. Further, fast guided filter has been employed to refine transmission map in place of soft matting. Results show that the proposed method gives better performance as image contrast, entropy and SSIM have increased and runtime cost has been significantly decreased. Problem of bluishness observed in classical method also has been reduced.