Fast motion estimation from pre-processing algorithm
Zhaozheng Hu · Ha'erbin gongye daxue xuebao · 2006
To reduce the computational complexity of motion estimation for video coding,a pre-processing algorithm for fast motion estimation was proposed.The algorithm firstly predicts the motion vectors distributions of current frame from their previous ones.Then the motion vectors with big distribution probabilities are pre-estimated.After the pre-processing step,a number of macro blocks' motion vectors have been estimated and only one searching point for these macro blocks is required.Thereby,the average number of searching points is reduced greatly.We give the theoretical analysis of the algorithm,which shows that it is extremely effective for the center-biased motion vectors distribution in most video sequences.In the experiments,the algorithm was applied in two classic algorithms: Three Step Search(TSS),and Diamond search(DS) algorithm.The results show that it can greatly reduce the average searching points,while retaining comparative sum of absolute difference(SAD) with those of TSS and DS.Therefore,it can greatly reduce the computational complexity of motion estimation,as well as the overall computational complexity for video coding.