A no-reference synthesized view quality assessment using statistical features in morphological multiscale domain
Dejan Bokan, Dragan Kukolj, Dragana D. Sandic-Stankovic · 2017
In this paper a no-reference quality assessment metric for DIBR-synthesized images is proposed. Proposed metric uses the statistical features of morphologically decomposed image subbands in order to estimate distortion level in synthesized images. Hoyer index as a measure of sparsity is used as a statistical feature. A Support vector regression is used to calculate quality score. The performance is evaluated using publicly available IRCCyN/IVC DIBR image database. Performance of the proposed measure is compared with commonly used full-reference and no-reference metrics. Experimental results show that proposed metric achieves good results in terms of correlation with human subjective quality perception.