Single Color Image Dehazing Based on Two Fast Variational Models

Abstract

Total Variation (TV) has been proven to effectively restrain the effect of noise in image processing. Multichannel TV (MTV) is proposed by extending TV model to adapt the case in color image processing. In order to effectively improve the performance of image restoration, we proposed to simultaneously consider denoising and dehazing by integrating MTV model with dark channel prior, called H-MTV model. With a better information preservation in detail, edge and texture of the image, nonlocal information is jointly considered with H-MTV model as NL-H-MTV model. Additionally, the two fast algorithms, namely dual bregman iteration and split bregman iteration, are respectively used for solving the H-MTV and the NL-H-MTV model, leading to a fast and accurate convergence. Experimental results on the several different images show that the performance of restoration using proposed methods are superior to those compared state-of-art methods.