Disocclusion Compensation Using Adaptive Template Matching for View Synthesis

Abstract

In a multi-view application, view synthesis is required to synthesize unavailable views using texture and depth information of the given views. In this paper, a compensation algorithm based on template matching for partial recovery of disocclusion regions is proposed. The shape of a template is dynamically adjusted, and a new reference mode which mimics pushing dominoes is also proposed. To find an appropriate template region, a histogram-based clustering approach is proposed to distinguish the background from the foreground objects. The improved performance is confirmed using subjective and objective evaluation methods.