A fast video stabilization algorithm based on block matching and edge completion

Abstract

Purpose of video stabilization is to register the frames of a video sequence with relative motions between each other to yield a stable video of higher perceptual quality. In this paper we focus on the problem of fast and robust video stabilization for the same scene based on temporal block matching. We use VoD principle to find the local motion vectors between adjacent frames, and then use statistical analysis to generate the global vibrant motion vector. After motion compensation, we further design an edge completion algorithm incorporating mosaicking and inpainting of neighbour frames, so as to reduce the impact of error propagation. Experimental results and comparative studies will be provided to justify the effectiveness of the proposed algorithm.