Improving distributed video coding side information by intelligently combining macro-blocks from multiple algorithms

The performance of distributed video coding (DVC) greatly relies on the quality of Side information (SI). This paper investigates a novel way of producing SI by intelligently combining macroblocks (MB) produced by two SI generation algorithms, namely higher-order piecewise temporal trajectory interpolation (HOPTTI) and adaptive overlapped block motion compensation (AOBMC). The two algorithms address the problem differently. HOPTTI attempts to improve the motion estimation using higher order trajectory interpolation while AOMBC addresses the blocking and overlapping problem caused by inaccurate block matching. By judiciously selecting when to incorporate AOBMC with HOPTTI, it would give a peak signal-to-noise ratio (PSNR) improvement in SI quality. Two switching mechanisms, which exploit the spatial-temporal correlation at the macro-block level, have been investigated and the RST-based intelligent mode switching (IMS) algorithm is found to produce enhanced SI quality. Experimental results show that the basic mode switching algorithm gives a PSNR improvement of up to 1.8dB in SI quality compared to using only HOPTTI. The more intelligent RST-based switching provides a further PSNR enhancement of up to 1.1dB for certain test sequences.