Variable grid size motion estimation using spatial transformations

The conventional BMA (block matching algorithm), which is the most popular video coding scheme assumes that a moving object in a scene undergoes only translational movements. To overcome this problem, general approaches to block matching motion estimation of the real objects by deforming the corresponding blocks in the picture have been proposed. However, conventional spatial transformation based motion estimation algorithms are not practical because of their heavy computational loads. We propose a variable grid size motion estimation method which is more efficient than conventional spatial transformation based methods and gives a better PSNR performance than the conventional BMA.