A two-pass optimal motion-threading technique for 3D wavelet video coding

Motion-threading is a novel technique that can efficiently incorporate motion information into 3D wavelet video coding. By utilizing a shape adaptive wavelet transform along with the motion-threading technique, 3D wavelet video coding can greatly outperform MPEG-4 for those sequences with simple motion. Since the motion estimation used in the threading process only considers the correlation of adjacent frames in signal magnitude, a large number of short threads are present in the existing motion-threading technique. As a result, the coding efficiency of the 3D motion-threading wavelet coding is far lower than that of MPEG-4 for those sequences with complex motion. This paper contributes a two-pass optimal motion-threading (TOMTH) technique for the 3D wavelet video coding. Both the correlation in signal magnitude and the number of threads are considered by using a new criterion of motion estimation. TOMTH effectively reduces the number of motion threads, meanwhile maintaining considerable accuracy in aligning frames. Experimental results show that the 3D wavelet video coding with the proposed TOMTH technique can achieve consistent good performance in various sequences.