Advanced template matching prediction using a motion boundary

Abstract

In this paper, an advanced template matching prediction is proposed to improve the coding efficiency gain. Conventional template matching prediction in joint exploration model (JEM) finds a motion vector for the current coding block by using adjacent template in reconstructed region although some coding blocks may have complicated motion coming from two moving objects or backgrounds. In these case, conventional template matching prediction mode could not be efficient. In order to overcome the complicated motion problem, we propose a new template matching prediction which applies two motion vectors to predict the coding block. In the proposed template matching prediction, a split mechanism is used to divide the coding block along the motion boundary. The proposed scheme is implemented on the JEM reference software 6.0 version and shows that maximum coding efficiency gain is up to 0.82% in luminance component.