Motion vector recovery for error concealment based on angular similarity

Abstract

This paper presents a motion vector (MV) recovery using the similarity of the angle of the corresponding surrounded MVs. The approach is based on the assumption that a group of macroblocks (MBs) which belongs to the same object and resides in the same region likely to move in the same direction. hence, those corresponded MVs that move in the same direction likely to have similar angle. As a result a lost motion vector can be estimated using a set of candidate MVs selected from the neighbouring MVs on the left, top-left, top, top-right, right, bottom-right, bottom and bottom-left. The experimental resuts for several test video sequences are compared with conventional error concealment methods and higher performance is achieved in both objective peak signal-to-noise ratio (PSNR) measurements and subjective visual quality.