Novel fractional pixel motion estimation algorithm using motion prediction and fast search pattern

Abstract

This paper presents a novel fractional motion estimation using motion prediction and fast search pattern for H.264/AVC. Fractional motion estimation improves coding efficiency and image quality. However, it requires additional computational overheads such as interpolation and fractional pixel search. To reduce fractional search computation, we propose a novel fast fractional search algorithm by combining the simplified adaptive search range and the mixed small diamond search pattern with the predicted factional motion vector. Compared with the fractional full search and the prediction-based directional fractional pixel search, the proposed algorithms can reduce up to 93% and 60% of fractional search points, respectively with the maximum PSNR loss less than 0.06dB. The proposed algorithm is quite suitable for mobile video applications for requiring low computational complexity.

A Fast Linear-Prediction Fractional-Pixel Search Algorithm

“…3 shows the interpolation relationship between the full-pixel positions (black), the 1/2-pixel positions (gray) and the 1/4-pixel positions (white) [9]. The fractional-pixel motion estimation can be used to improve the inter-frame prediction for video quality [7]. However, it requires large computations, such as fractionalpixel (1/2-and 1/4-) search algorithm and interpolation process [7].…”

“…Therefore, if the integer-pixel motion vector is selected as the final motion vector, the fractionalpixel search computations can be skipped. In this paper, a new FLFS method is proposed to determine whether the fractional-pixel search should be performed or not [7] [8].…”

“…One observation is that if the minimum sum of absolute difference (SAD) obtained at fractional-pixel accuracy is larger than that at integer-pixel accuracy, the motion vector on the integer-pixel is selected as the final result and the fractional-pixel search for this block is regarded as ineffective. Otherwise, the fractional-pixel search is regarded as effective and the motion vector on the fractional-pixel is selected as the final result [7] [8]. Table I shows the effective fractional-pixel search ratio for some test video sequences.…”

“…The fractional-pixel motion estimation can be used to improve the inter-frame prediction for video quality [7]. However, it requires large computations, such as fractionalpixel (1/2-and 1/4-) search algorithm and interpolation process [7]. One observation is that if the minimum sum of absolute difference (SAD) obtained at fractional-pixel accuracy is larger than that at integer-pixel accuracy, the motion vector on the integer-pixel is selected as the final result and the fractional-pixel search for this block is regarded as ineffective.…”