Image dehazing based on structure preserving

Abstract

Restoring the scene radiance from degraded image is a challenging problem in computer vision. This paper proposes a novel dehazing method from single image based on structure preserving. Different from most existing methods, the structure information is considered sufficiently for visibility enhancement. To start with, the initial airlight is derived by filtering the minimum channel of hazy image, meanwhile the structure information of the minimum channel is extracted as the reference image. Then, the initial airlight is refined with the reference image by guided joint bilateral filter, which makes the scene radiance and depth more naturally. Finally, the scene radiance is restored by solving the atmospheric attenuation model. Specifically, an improved method based on quad-tree subdivision is presented to obtain an accurate atmospheric light. We verify the effectiveness and feasibility of the proposed method on series of real hazy images. Experimental results indicate that the proposed method can achieve comparable or even better dehazing results than several well-known methods in view of subjective and objective evaluations.