| Hi-index | 0.00 |
Block matching algorithms (BMA) have beenwidely used as standard motion estimation (ME)techniques in video coding practice.In this paper,we present an iterative but analytical motionestimation algorithm with attractive computationalefficiency.Our work can be viewed as generalizationof the celebrated work of Lucas and Kanade[1] on iterative image registration into the waveletspace.We will demonstrate that block-based motionestimation can be performed efficiently in a coarse-to-finefashion by propagating the estimate acrossthe scale in the wavelet space.Our solution oftentakes a fraction of the computations required by theconventional BMA methods and reches comparableMSE performance of prediction residues (fractional-pelmotion accuracy is achieved due to the analyticalnature of the proposed scheme).In spite of theiteration in the proposed scheme, it converges fastespecially with a good initial estimate passed onfrom the low resolution.Our new ME scheme isparticularly useful to the speed-up of implementing anew class of wavelet-based scalable video coders.