A Triangular Mesh Motion Compensation Method in Wavelet Domain

The block-based motion estimation(ME)and motion compensation(MC)are not suitable for rotation, expansion and contraction motion models in image sequences. The mesh-based ME/MC is helpful for dealing with this case, but it is hard to be used in the pyramid structure of DWT, especially in the low-resolution subbands. This paper proposes a wavelet domain triangular mesh motion compensation method(WDTM-ME/MC). The nodes of triangular mesh are chosen in redundant discrete wavelet transform(RDWT)domain of reference frame, the mesh nodes numbers are adjusted according to the grade of motion included in image sequences. Delaunay triangulation is carried out and the mesh nodes forward motion tracking implements the ME function. The mapping function between mesh elements of current frame and reference frame outputs the prediction of current frame in RDWT domain, implements the MC function. The procedure of transforming prediction from all phase RDWT subbands into spatial implicitly includes a multihypothesis prediction. The proposed WDTM-ME/MC method improves the efficiency of ME/MC by making use of the spatial shift invariance of RDWT, mesh-based ME/MC’s good application for multi motion modes and multihypothesis prediction. The experimental results prove that the WDTM-ME/MC method can improve the efficiency of ME/MC.