No-reference stereoscopic image quality assessment based on saliency-guided binocular feature consolidation

摘要

Different from traditional methods depending on the procedure of intermediate `cyclopean' view construction, a novel framework based on saliency-guided multi-scale feature consolidation for stereoscopic image quality assessment is proposed. For quality representation, the underlying features are extracted from three aspects: (i) global natural statistics features, (ii) local spatial and spectral entropy features and (iii) the kurtosis and skew of disparity distribution. Then the binocular features are consolidated by a saliency-guided weighted process. Finally, a machine learning technique of support vector regression is used for objective quality mapping. Experimental results demonstrate the promising performance of the proposed method.