Fast Multi-Hypothesis Motion Compensated Filter for Video Denoising

  • Authors:

  • Liwei Guo;Oscar C. Au;Mengyao Ma;Zhiqin Liang

  • Affiliations:

  • Thomson Corporate Research, Princeton, USA 08540;Department of Electronic and Computer Engineering, Hong Kong University of Science and Technology, Clear Water Bay, Hong Kong;Department of Electronic and Computer Engineering, Hong Kong University of Science and Technology, Clear Water Bay, Hong Kong;Harmonic Inc., Hong Kong, Hong Kong

  • Venue:
  • Journal of Signal Processing Systems
  • Year:
  • 2010

Quantified Score

Hi-index 0.00

Visualization

Abstract

Multi-Hypothesis motion compensated filter (MHMCF) utilizes a number of hypotheses (temporal predictions) to estimate the current pixel which is corrupted with noise. While showing remarkable denoising results, MHMCF is computationally intensive as full search is employed in the expectation of finding good temporal predictions in the presence of noise. In the frame of MHMCF, a fast denoising algorithm FMHMCF is proposed in this paper. With edge preserved low-pass prefiltering and noise-robust fast multihypothesis search, FMHMCF could find reliable hypotheses while checking very few search locations, so that the denoising process can be dramatically accelerated. Experimental results show that FMHMCF can be 10 to 14 times faster than MHMCF, while achieving the same or even better denoising performance with up to 1.93 dB PSNR (peak-signal-noise-ratio) improvement.