Single Image Dehazing Using Local Detail Enhancement

Abstract

Most existing single image dehazing algorithms require the estimation of atmospheric light using simple procedures based on error-prone assumptions. In this letter, a new dehazing method based on local detail enhancement is proposed to estimate atmospheric light and transmission map by considering local detail enhanced images as quasi-haze-free images. The proposed method is based on the decomposition model that interprets the transmission map as a base layer and the haze-free image as a detail-like layer. From a hazy image, local detail information is extracted using the local detail enhancement approach based on an edge preserving filter. Then, atmospheric lights and transmission maps for each color channel are estimated using Koschmieder's law. The transmission map is refined using the dark channel prior and a guided filter. The experimental results show the proposed method can remove the layer of haze effectively and produce better performance than existing dehazing methods.