Image dehazing method quality assessment

In recent years, responding to subjective assessment results in assessment methods by targeting the quality of dehazing algorithms has become common, but this lacks quantitative description. However, existing objective quality assessment methods and subjective assessment have been inconsistent, and the two sometimes diverge. Therefore, to improve the objective quality assessment performance of a dehazing method, a full reference dehazing method quality assessment based on artificially synthesized images is proposed here. First, a synthetic image dataset is established that includes reference haze-free, synthetic hazy, and dehazed images obtained by using eight state-of-the-art dehazing algorithms on these synthetic hazy images. Second, we classify quality problems that might be introduced by the dehazed images. Third, by combining clarity-related features and existing objective methods of image quality assessment, a dehazing method quality assessment is proposed through mutual integration of image visibility, structural similarity, and color recovery. In the synthetic image dataset, this paper's method is compared with existing image quality assessment methods for experiments. The experimental results showed that, for the synthetic image dataset, the proposed method performed optimally in SRCC, PLCC, and RMSE metrics. The consistency of this paper's method with subjective assessment was better, which was more favorable to support research on dehazing algorithms. © 2022, Science Press. All right reserved.