Precise sub-pixel estimation on area-based matching

Abstract

Area-based matching is a common procedure in various fields such as image-based measurements, stereo image processing, and fluidics. Sub-pixel estimation using parabola fitting over three points with their similarity measures is also a common method to increase the resolution of matching. However, few investigations or studies concerning the characteristics of this estimation have been reported. In this paper we have analyzed the sub-pixel estimation error by using an approximate image function and three kinds of similarity measures for matching. The results illustrate some inherently problematic phenomena such as so called "pixel-locking". In addition to this, we propose a new algorithm to greatly reduce sub-pixel estimation error. This method is independent from the similarity measure and quite simple to implement. The advantage of our novel method is confirmed through experiments using three different types of images.