A novel non-reference image quality assessment algorithm

摘要

Due to the poor performance of many non-reference evaluation algorithms for the TID2013 image library, it is a great challenge to evaluate non-reference image quality for such library. In this paper, a novel non-reference image quality assessment(NR-IQA) algorithm based on wavelet energy, structure and texture features and color signature is proposed. The wavelet energy is calculated by combining visual saliency map and four level wavelet transform. The textural and structural map are obtained by Rudin-Osher-Fatemi algorithm while the features of these are extracted by local binary pattern (LBP) method. Two types of characteristic parameters of distorted image are analyzed, among which one is acquired in gray-scale image and the other is obtained in two color channel (Cb and Cr). Finally, Back Propagation artificial neural network and Radial Basis Function neural network are used to learn the mapping between feature space and subjective opinion scores. Experimental results on two benchmark image quality databases show that the proposed method has highly competitive performance in the state-of-the-art NR-IQA theories, especially in TID2013 IQA database.