Fast Binary Block Matching Motion Estimation using Efficient One-Bit Transform

Binary block matching algorithm employs one-bit transform (1BT) to transform video sequences from full resolution representation (8-bit/pixel in general) to 1-bit/pixel bit-plane, then performs block matching motion estimation in bit-plane to save computations. Unfortunately, current 1BT algorithms either have bad output motion vectors or are computational intensive. In this paper, a novel fast and efficient 1BT algorithm called the overlap-windowed thresholding algorithm (OWT), and a new fast binary block matching algorithm (FBBMA) based on OWT are proposed. Besides OWT, there are three other novelties in FBBMA, namely fast binary block matching, adaptive central-search-point prediction and center-biased search order, in which significant modifications are contained. Four more existing techniques are employed in FBBMA to further improve performance. Experimental results show the superiorities of all the proposed algorithms and modifications. With similar quality of predicted sequence, the proposed OWT algorithm performs 37 times faster than the filter thresholding method, which gives the best quality among current 1BTs. FBBMA significantly outperforms over several other fast motion estimation algorithms, including the diamond search algorithm that was recently adopted in MPEG-4, in terms of both speed up ratio and quality of predicted sequence. In addition, the output motion vectors of FBBMA need fewer bits to be encoded than most if not all other ME algorithms.