Prediction-Based Flexible Triangle Search Algorithm for Block Based Motion Estimation

This paper presents a technique that predicts the initial search direction for the flexible triangle search algorithm (FTS), which was introduced in previous work by the authors. The FTS algorithm is a fast block-matching algorithm for block-based motion estimation. In the FTS, a search triangle is used to find the best matching blocks between two frames through successive iterations. During the search, the triangle changes its direction and size using reflection, expansion, contraction, and translation operations. These operations provide the triangle with the necessary flexibility to perform coarse or fine search and to locate the best matching blocks while checking fewer search positions compared to most other search algorithms. Analysis of the FTS behavior showed that the proper selection of the starting triangle for the search reduces the required number of block-matching evaluations by directing the search earlier in the direction of the minimum. In this paper, a prediction-based FTS, the PFTS, is introduced. In the PFTS, a prediction step is added to obtain the initial triangle for the search. Simulation results indicate that the PFTS requires a smaller number of block matching operations than that of the FTS. In addition, the compression ratio was improved slightly while the visual quality of the reconstructed sequence remained the same compared to the FTS