Motion estimation based on region prediction for fixed pattern algorithms

Motion estimation plays an important role in video compression. The most accurate strategy for motion estimation is the Full Search algorithm (FS) which checks all possible candidate motion vectors to find the optimal result. However, its significant computational load is a major drawback for implementation. Many algorithms have been introduced to reduce the computational complexity of FS algorithm. Most of them use fixed search patterns in their first step. They divide the search window into four regions and choose some blocks in each region to calculate Sum of Absolute Difference (SAD). Motion vectors of neighboring blocks are usually well-correlated with the current block´s motion vector. Therefore, they can be used to predict the region in which the best matched block is located. So there is no necessity to search other blocks in the other regions, speeding up the whole algorithm. In some cases PSNR has increased because the prediction usually prevents the algorithm from falling into the local minima.