Tracking performance of several combinations of common evaluation metrics and sub-pixel methods

Abstract

Motion estimation in a series of consecutive images is used in a variety of areas, e.g. video compression and investigation of tissue characteristics and organ function in medical images. Several methods exist both for estimating motions on a pixel level, e.g. block-matching in which two blocks in consecutive images are compared by an evaluation metric, and on a sub-pixel level. In this paper, we have evaluated the tracking performance of all combinations between three evaluation metrics and eight sub-pixel estimation methods. The tracking performance of a sub-pixel method varies depending on the evaluation metric used. This indicates that a reported tracking performance for a sub-pixel estimation method can be significantly different when combined with another evaluation metric. Also there is a large variation in the time needed for the motion estimations depending primarily on the sub-pixel method used but also on the evaluation metric.