Advanced motion threading for 3D wavelet video coding

This paper presents an advanced motion threading technique for improved performance in 3D wavelet coding. First, we extend an original motion threading idea of ours to a liftingbased implementation. Methods for processing many-to-one pixel mapping and non-referred pixels and for enabling fractional-pixel alignment in motion threading are proposed to reduce the wavelet boundary effects. Second, we devise an advanced motion threading technique, in which one set of motion vectors is generated for each temporal layer of wavelet coefficients for temporal scalability. In order to reduce the motion overhead information, especially at low bit rates, several correlated motion prediction modes at the macroblock level are defined to exploit the intra/inter layer correlation in motion vector coding. Finally, rate-distortion optimization is utilized in motion estimation to select the best motion prediction mode for each macroblock. With the new motion threading technique, we are able to achieve 1.5-6.0 dB gain in average SNR in 3D wavelet coding over our previous implementation of motion threading.