A Low-complexity Wavelet Based Algorithm for Inter-frame Image Prediction

Abstract

In this paper, a novel multi-resolution variable block size algorithm (MRVBS) is introduced. It is based on: (1) Using the wavelet components of the seven sub-bands from two layers of wavelet pyramid in the lowest resolution; (2) Performing a block matching estimation within a nine-block only in each sub-band of the lower layer; (3) Scaling the estimated motion vectors and using them as a new search center for the finest resolution. The motivation for using the multi-resolution approach is the inherent structure of the wavelet representation. A multi-resolution scheme significantly reduces the searching time, and provides a smooth motion vector field. The approach presented in this paper providing an accurate motion estimate even in the presence of single and mixed noise. As a part of this framework, a comparison of the Full search (FS) algorithm, the three-step search (TSS) algorithm and the new algorithm (MRVBS) is presented. For a small addition in computational complexity over a simple TSS algorithm, the new algorithm achieves good results in the presence of noise.