a significant loss of the picture quality. Moreover, the proposed technique can also be applied to almost all the ex-An adaptive early jump-out technique for speeding up the block-based motion estimation is proposed. By using the new isting fast motion estimation algorithms, such as the threetechnique, we can speed up the full range search several times step search, the conjugate direction search, and the interwithout losing the picture quality significantly. The proposed block correlation search, and still gains the improvement technique can also be embedded in almost all the existing fast on the computational speed. Our technique has been tested motion estimation algorithms to speed up the computation on the H.261, the H.263, and the MPEG-I codecs, and the further. Since the proposed technique can be embedded in the coding speed does improve significantly.
existing motion estimation algorithms, it can be applied to
This paper is organized as follows. Section 2 briefly re-
all the standard video codecs, such as the MPEG coder, views the block-based motion estimation algorithm and the and improve the coding speed of such codecs significantly. early jump-out techniques. The proposed adaptive early Our technique has been tested on the H.261, the H.263, and jump-out technique is presented in Section 3. Section 4 the MPEG-I codecs, and the coding speed does improve significantly. ยฉ 1997 Academic Press describes the order for searching and the order for computing the match error. The experimental results are shown in section 5. Finally, the conclusion and future work are 1. INTRODUCTION stated in Section 6. Motion estimation is one of the major parts of video 2. PRELIMINARY REVIEWS coding standards [1, 2] and multimedia systems. The high In this section, we briefly review the motion estimation computational cost of motion estimation is the major probalgorithms and the early jump-out techniques. lem of the coders. Many fast motion estimation algorithms, such as the three-step search [3], the 2-D logarithmic algo-2.1. Block-Based Motion Estimation Algorithms rithm [4], the conjugate direction search [5], the successive elimination algorithm [6], and the inter-block correlation Most video coding standards use motion compensation algorithms [7, 8], have been proposed for solving the comto exploit temporal redundancy in the video. Decoders putation problem of motion estimation.
construct a predicted block of pixels from pixels in a pre-In this paper, the adaptive early jump-out technique for viously transmitted image. Motion within the images usumotion estimation algorithms is proposed. The early jumpally implies that the pixels in the reference image, usually out (EJO) technique has been applied for speeding up the previous image, will be in a different position from the pixels in the current image, and the displacement, motion corner detection and template matching in [10][11][12] and has achieved very impressive results. However, since the vector, is estimated by the video encoder and encoded in the bitstream. The predicted image block is usually a good characteristic of motion estimation is quite different from those of corner detection and template matching, both estimate of the current image block, and it is usually more efficient to transmit the motion vector plus the difference the mathematical early jump-out models [10,11] and the trained early jump-out sequence [12] are unsuitable for between the predicted block and the current block than to transmit a description of the current block by itself. motion estimation. An adaptive early jump-out threshold sequence training algorithm is proposed. The adaptive
The motion estimation of the video encoder determines the best motion vectors. Using a search strategy the en-early jump-out technique has the capability of learning the characteristic of the image sequence on-line. The experi-coder attempts to match the pixels in an image block with those in the reference image. The vector corresponding to mental results show that the adaptive early jump-out technique can speed up the full range search algorithm without the best match is reported after the search is completed.