Image Quality Assessment Based on Distortion-Aware Decision Fusion

Generic image quality (IQ) metrics based on individual features are not capable of making accurate prediction across different distortion types. In this paper, we propose a two-stage scheme to overcome this limitation. At the first stage, the image distortion type is predicted by support-vector classifiers. At the second stage, decision-level fusion of three existing IQ metrics are conducted based on the k-nearest-neighbor (k-NN) regression where the acquired distortion-type knowledge is employed. When evaluated on the largest publicly-available IQ database which involves a large variety of distortion types, the proposed approach demonstrates impressive accuracy and robustness.