Study On Coding Of Stereo Image Sequences

Along with rapid development in stereo display technology, stereo images/sequence would take place of two-dimensional (2D) images/sequence in near future. While the information amount of stereo image/sequence is two folds to that of 2D images/sequence, a series of efficient coding schemes are required. Conventional coding schemes endure complex computation, low coding efficiency, and poor practicability, so that this paper proposes a few coding schemes based upon the inherent feature of stereo images.This paper proposed a pseudo disparity-based stereo image coding according to the different meanings of disparity vectors between stereo image coding and stereovision. Since the disparity vectors in this scheme are not displacement of correspondent points in left and right images, they only denote the binocular redundancy inherent in stereo pair. The experimental results show that the proposed scheme achieves the reasonable reconstruction under the bit rate of 0.4bpp. Moreover, the computational cost is reduced too much due to the simplified block-based stereo matching approach. Furthermore, the proposed scheme can achieve better reconstruction than JPEG2K used to encode independently under the identical bit rate. In addition, the encoding time of the proposed scheme is only half of that of JPEG2K.In order to improve the reconstruction of boundary of objects, this paper presents a variable-block pseudo-disparity stereo coding scheme. Whether the block image is partitioned into 4x4 sub-blocks lies on the summation of absolute difference (SAD) of grey pixel values. Supposed that SAD is more than the pre-designated threshold value, and then the block image will be done that. The experimental results demonstrate the efficiency of the proposed scheme. Therefore, the proposed scheme will push the speed of stereo imaging applications.In addition, this paper reviews the existing coding schemes for stereo image sequences. Since stereo image sequences include not only temporal redundancy but also binocular redundancy inherent in left channel and right channel, existing algorithms remove the temporal redundancy with motion vectors and binocular redundancy with disparity vectors. Even though three-mode joint prediction is a high-performance coding scheme, two bits used for selecting three modes are not efficient in that increment one bit is only used for one mode. In order to further improve the coding efficiency, this paper proposed a flexible and efficient coding scheme for stereo image sequence. The proposed scheme adopts variable-weight joint prediction modes coupled with fast estimation of motion and disparity vectors. The experiment results indicate that the proposed scheme can achieve more flexible and efficient coding performance and faster speed than existing schemes.