Single Image Dehazing Method Based on Gradient Matrix and Ternary Parameter Search

Chuheng Chen, Juping Liu, Kai Chen, Mingye Ju

Abstract: Hazy images typically suffer from low contrast and poor visibility, which influences the performance of many computer vision systems. In this paper, a single image dehazing method called DG T is proposed. First of all, by employing the dark channel prior, atmospheric light and rough transmission map can be obtained. In subsequence, via utilizing the gradient matrix, the object details in the images can be enhanced. Finally, built on a minimum information loss principle, a ternary search algorithm is employed to determine the related parameters, and thus effectively restore the hazy images. Different from the previous works, DGT only needs to determine one unknown parameter with the cost of logarithmic time complexity. Experimental results show the superiority of the proposed method over state-of-the-art dehazing methods.