Deformable Block Motion Estimation and Compensation in the Redundant Domain
Ming Yu, Yue Zhao, Yingchun Guo, Feng-quan Zhang · 2008
Because the redundant discrete wavelet transform (RDWT) is shift invariant, the motion estimation algorithms in the redundant wavelet domain have good performance. And, deformable block matching algorithm (DBMA) not only can reduce 'block effect' which is brought by traditional block matching algorithm but also can have an effective prediction for the non-translational motion. So, this paper mainly focus.es on deformable block motion estimation and compensation in the redundant discrete wavelet transformation domain (RDWT-DBMA). The motion estimation is done in the biorthogonal 9/7 redundant wavelet domain, and four-nodes-based model is used in nodal-based deformable matching algorithm. Experimental results show that RDWT-DBMA has a better subjective and objective image quality than full search algorithm (FS) which has the best precision in block matching algorithm (BMA), especially for the sequence which has fast motions, and the average PSNR of the image can be improved 3.17dB.