Regions of interest aware versatile video coding for 360° videos

Widespread adoption of virtual reality applications continuously urges for better compression of 360° videos due to the large volume of the content. Since a user’s observation is restricted to a limited number of viewports in 360° videos, identification of such regions prior to encoding and adaptively encoding such contents according to the identified regions can both improve the perceptual quality and reduce the bitrates. In response, this paper proposes a novel Region of Interest (ROI) aware pre-processing scheme for 360° video coding. To this end, first, the user observed viewports are estimated by applying the bidirectional optical flow technique. Later, multiple Quality Emphasis Regions (QERs) are fused in order to compose the ROI suitable for encoding. Finally, the viewports are adaptively encoded using an ROI aware video encoder. Experiments in the context of Versatile Video Coding (VVC) Test Model 7.0 show that the proposed method can improve intra-coding compression efficiency up to 10% at the same perceptual video quality.