On the performance of objective metrics for omnidirectional visual content

Abstract

Abstract-Omnidirectional image and video have gained popularity thanks to availability of capture and display devices for this type of content. Recent studies have assessed performance of objective metrics in predicting visual quality of omnidirectional content. These metrics, however, have not been rigorously validated by comparing their prediction results with groundtruth subjective scores. In this paper, we present a set of 360-degree images along with their subjective quality ratings. The set is composed of four contents represented in two geometric projections and compressed with three different codecs at four different bitrates. A range of objective quality metrics for each stimulus is then computed and compared to subjective scores. Statistical analysis is performed in order to assess performance of each objective quality metric in predicting subjective visual quality as perceived by human observers. Results show the estimated performance of the state-of-the-art objective metrics for omnidirectional visual content. Objective metrics specifically designed for 360-degree content do not outperform conventional methods designed for 2D images.

Bridge the Gap Between VQA and Human Behavior on Omnidirectional Video

“…For objective VQA on omnidirectional video, there have been several approaches [35,38,41,45,49,50] that advance the metric of PSNR by considering the sample density of the map projection in omnidirectional video. Some applied weight allocation in the calculation of PSNR, e.g, weighted-to-spherically-uniform PSNR (WS-PSNR) [35] and area weighted spherical PSNR (AW-SPSNR) [45]; others resampled the content into a point-uniformed projection, e.g., S-PSNR [49] and PSNR in Craster parabolic projection (CPP-PSNR) [50].…”

“…Some applied weight allocation in the calculation of PSNR, e.g, weighted-to-spherically-uniform PSNR (WS-PSNR) [35] and area weighted spherical PSNR (AW-SPSNR) [45]; others resampled the content into a point-uniformed projection, e.g., S-PSNR [49] and PSNR in Craster parabolic projection (CPP-PSNR) [50]. In [38,41], the performance of the above objective VQA approaches was compared by measuring their correlation with the subjective quality scores. Unfortunately, none of the above VQA approaches considers human behavior on viewing omnidirectional video, which significantly influences the quality of experience (QoE) [17,37].…”

“…Therefore, it is necessary to study on visual quality assessment (VQA) for omnidirectional video. Most recently, several subjective VQA approaches [20,28,30,31,39,40,47,51] and objective VQA approaches [35,38,41,45,49,50] have been proposed for either omnidirectional image or video. Among these approaches, the impact of some factors on the quality of omnidirectional video were studied, such as display types, coding schemes and sample uniformity under different map projections.…”

A simple method to obtain visual attention data in head mounted virtual reality

“…The data used in the present work has been obtained during a subjective quality evaluation experiment [14] on omnidirectional images. For the current study we selected only the head direction tracks recorded from unimpaired stimuli.…”

Graph-Based Detection of Seams In 360-Degree Images

“…Although approaches to quantify the overall visual quality of 360-degree images and videos have been proposed in literature [5][6][7][8][9][10][11], they consider uniformly distributed distortions due to the processing of content in ER domain, rather than localized distortions that might occur when processing the signal in other domains, such as the CM one [2].…”