Embedded coding of the motion-compensated 3-D wavelet coefficients by concatenating spatial and temporal orientation trees

This paper focuses on the issue of coding of the spatiotemporal 3-D wavelet coefficients. Experimental examination found that there are temporal dependencies among the wavelet coefficients at different temporal decomposition levels. Temporal orientation trees are introduced to exploit these dependencies to improve the coding efficiency. Statistical properties of the wavelet coefficients are measured using test video sequences. 3-D orientation trees that are built by spatial orientation trees followed by temporal orientation trees are exploited to magnitude-order the spatiotemporal 3-D wavelet coefficients. The experimental results confirmed that the coder with temporal orientation trees outperformed the coder without them. The coding gain could be up to 0.5 dB.