An effective performance ranking mechanism to image dehazing methods with psychological inference benchmark

Abstract

In this paper, we proposed a novel quality assessment framework for the performance ranking of image dehazing algorithms by employing prior features and radial basis function (RBF-based) classifier. First, we formulate the evaluation problem within a novel comparison framework by using classification methods. Second, prior information is used to extract the inherently feature of hazy image and these feature values are normalized with psychological inference benchmark (PIB) to eliminate the cognitive bias of individuals. Finally, a cost-compensation classification network is cyclically utilized to rank the image dehazing performance until the iteration ends and updated the PIB in every loop. Experiments show that the proposed method is more effective for evaluating the image dehazing performance than the existing blind image quality assessment methods, and the evaluation results correlate well with human judgments of image quality.