Motion estimation via cluster matching
D.P. Kottke, Ying Sun · IEEE Transactions on Pattern Analysis and Machine Intelligence · 1994
A new method for estimating displacements in computer imagery through cluster matching is presented. Without reliance on any object model, the algorithm clusters two successive frames of an image sequence based on position and intensity. After clustering, displacement estimates are obtained by matching the cluster centers between the two frames using cluster features such as position, intensity, shape and average gray-scale difference. The performance of the algorithm was compared to that of a gradient method and a block matching method. The cluster matching approach showed the best performance over a broad range of motion, illumination change and object deformation.>