EXTENDED ADAPTIVE FILTERING FOR MOTION COMPENSATED PREDICTION(International Workshop on Advanced Image Technology 2006)

概要

論文の詳細を見る
Most standardized video coding systems such as MPEG-1/2/4 and H.263/264 are based on motion compensated prediction with fractional-pel accuracy. In these systems, samples at fractional-pel positions are interpolated by an invariant filter whose coefficients are pre-determined according to the position. Wedi has recently proposed a number of adaptive interpolation filtering methods for motion compensation in order to improve coding efficiency. In these methods, the interpolation filter coefficients are designed in response to the changing statistical properties of image signals, and the samples at fractional-pel positions are interpolated adaptively. In this paper, we propose an improved motion compensated prediction method for samples at full-pel positions, which are located at sampled positions. This method works as adaptive filter for samples at full-pel positions. In addition, it makes compensation for luminance transition in fade-in/fade-out scene, etc. The experimental results for several test sequences show a PSNR gain up to 0.27 dB in comparison with H.264/MPEG-4 AVC. For a sequence including a fade-in scene, PSNR gains of 2.48 dB and 0.26 dB are obtained in comparison with H.264/MPEG-4 AVC with and without weighted prediction, respectively.
社団法人電子情報通信学会の論文
2006-01-03