Inter view motion vector prediction in multiview HEVC

Döne Bugdayci Sansli, Kemal Ugur, Miska M. Hannuksela, Moncef Gabbouj

Abstract: In multiview coding, motion information of different views forming the multiview video is correlated. The multiview extension of High Efficiency Video Coding (MV-HEVC) that is currently under development is defined to allow multi-view coding with minimal changes to HEVC. This is realized by simply adding the reconstructed base view picture to the reference picture list of the dependent views. MV-HEVC specification is restricted from low-level operations with motion information, thus redundancy in motion information between views is not utilized. This paper proposes to use the temporal motion vector prediction mechanism of HEVC to enable the prediction of motion vectors from the inter-view reference picture; similar to the concept of Motion Field Mapping mechanism in the scalable extension of HEVC. In order to compensate for the disparity, it is also proposed to signal a global disparity value for motion field adjustment. Global disparity is estimated at the encoder side for each slice and signaled to the decoder side at the slice header. Simulations conducted in the MV-HEVC configuration of 3D-HTM v6.0 show that the proposed method brings 1.1% of dBR for view 1 and 0.8% for view 2 which results in 0.6% gain on average of all views.