A ‘Complete Blind’ No-Reference Stereoscopic Image Quality Assessment Algorithm

Abstract

We propose a complete blind no-reference (NR) image quality assessment algorithm for assessing the perceptual quality of natural stereoscopic (S3D) images. Towards this end, we have generated an intermediate image from the left and right views, and hypothesize that the perceived quality of the S3D view close to that cyclopean image. We perform multi-steerable decomposition on cyclopean images and we compute the naturalness image quality evaluator (NIQE) score [1] and entropy score from each subband. Finally, the primitive quality scores of steerable subbands are pooled to obtain the overall perceptual quality score of an S3D image. The proposed algorithm is evaluated on the LIVE Phase I [2] and LIVE Phase II [3] stereoscopic image datasets and demonstrates its robust performance on both the datasets and across distortions. The proposed algorithm, which is a ‘complete blind’ model (neither requires pristine S3D images nor requires training on human opinion scores), is called the Multi-Orient NIQE based 3D image quality evaluator (MO-NIQE).