A motion-estimation architecture based on band-matching

Abstract

An algorithmic specific architecture has been developed based on the newly introduced band-matching algorithm (BDMA) for computing motion vectors. The BDMA offers better performance than both the fixed size and variable size block matching algorithms as it uses less vectors and only processes the relevant parts of the image instead of the whole image. The main modules of the BDMA architecture are the encoder and decoder. The memory in the encoder is classified into three types: storing images, storing the image of difference, storing the bands. The first memory is arranged in parallel banks which will enhance the performance considerably. In the second memory, registers are used for temporary storage during filtering. The third memory is used for storing the bands. A single code book is employed to identify the type of edges.