Pixel-Level Dehazed Image Quality Assessment Based on Dark Channel Prior and Depth

Image dehazing has attracted wide attention as a promising technology to recover the hazy image to a clear image. As many algorithms are proposed to conduct single image dehazing, the way of how to evaluate the dehazed image quality effectively has become an important issue in the field of image dehazing. In this paper, we propose a novel reduced-reference Pixel-level Dehazed Image Quality Assessment (PDIQA) method by incorporating dark channel information. The proposed method can accurately measure the effect of image dehazing, and its pixel-by-pixel dehazing evaluation can guide the dehazing algorithms (DHA) to achieve the removal of residual haze. To evaluate different dehazing algorithms objectively, we firstly build a real-world dataset including 15 real-world hazy images combined with 75 dehazed results from 5 different state-of-the-arts dehazing algorithms. Then, we carry out a subjective quality evaluation study on this dataset. A large number of experiments verify the effectiveness of the proposed method.