A diamond search windowbased adaptive search range algorithm

Abstract

In this paper, a novel diamond search window (SW) based full search algorithm is proposed. To dynamically adjust the size of the diamond search window, a frame level adaptive search range (SR) algorithm is presented. Laplace distribution is used to model the optimal motion vector difference (MVD) distribution within a frame and the model parameters are estimated by maximum likelihood estimation (MLE). The data used for MLE is also carefully studied. With the estimated distribution model, proper SR is obtained dynamically frame by frame. Experiment results show that a diamond SW outperforms the traditional squared SW. Also, the proposed adaptive search range algorithm properly adjusts the diamond SW size. The proposed diamond SW based adaptive search range (ASR+DSW-FS) algorithm obtains the same or even better performance compared with conventional full search algorithm, while significantly reduces the computational complexity.