No-Reference Quality Assessment for 360-Degree Images by Analysis of Multifrequency Information and Local-Global Naturalness

360-degree/omnidirectional images (OIs) have received remarkable attention due to the increasing applications of virtual reality (VR). Compared conventional 2D images, OIs can provide more immersive experiences consumers, benefiting from higher resolution and plentiful field views (FoVs). Moreover, observing is usually in a head-mounted display (HMD) without references. Therefore, an efficient blind quality assessment method, which specifically designed for 360-degree urgently desired. In this paper, motivated by characteristics human visual system (HVS) viewing process VR content, we propose novel effective no-reference omnidirectional image (NR OIQA) algorithm MultiFrequency Information Local-Global Naturalness (MFILGN). Specifically, inspired frequency-dependent property cortex, first decompose projected equirectangular projection (ERP) maps into wavelet subbands using discrete Haar transform (DHWT). Then, entropy intensities low-frequency high-frequency are exploited measure multifrequency information OIs. addition considering global naturalness ERP maps, owing browsed FoVs, extract natural scene statistics (NSS) features each viewport as local naturalness. With proposed measurement local-global measurement, utilize support vector regression (SVR) final regressor train evaluation model quality-related ratings. To our knowledge, method that combines Experimental results on two publicly available OIQA databases demonstrate MFILGN outperforms state-of-the-art full-reference (FR) NR approaches.