Efficient motion re-estimation method based on k-means clustering for spatial resolution reduction transcoding

In this paper, we present a spatial resolution reduction transcoding architecture for H.264/AVC with a low-complexity compensation technique. A fast motion vector re-estimation based on a k-means clustering algorithm is proposed for efficiently determining a motion vector and a reference frame. A motion vector refinement technique is used to further improve the rate-distortion performance without additional high complexity. The proposed method significantly reduces the motion vector estimation complexity up to 10 times while the loss of rate distortion performance is negligible with a decrease of 0.1dB in BD-PSNR and 3% increase in the BD-RATE.