Improving Histogram-Based Image Registration in Video Sequences through Warping

Abstract

This paper presents two computationally efficient dynamic-time warping algorithms for image registration in video sequence through histogram based image segmentation. The key idea is to warp the histogram in an input frame to create an approximation of a reference frame. Any histogram based thresholding method can then be applied to create consistent regions in both the input and the approximated reference frames. Experiments of the proposed algorithm are used to demonstrate that more consistent matches can found after thresholding.