DCT-based embedded coding scheme for stereo image

Recently, embedded stereo image coders have attracted a lot of attention, not only because of limited transmission bandwidth, but because the human visual system can perceive a sharp 3D image even when one of the images is coded as a lower quality image. We propose a novel DCT-based embedded stereo coder. Firstly, an adaptive Delaunay triangulation (DT) mesh-based disparity estimation (DE) algorithm is used for obtaining disparity vectors. Then, we utilize a Markov random field (MRF) model to optimize the disparity vectors further, which results in a smoother disparity field. Moreover, an improved DCT-based bit-plane coder providing an embedded bit stream is used to encode the residual image. Different from the conventional VLC-table bit-plane coding algorithm, this algorithm imposes a context-based arithmetic coding algorithm, whose context models are specially designed by studying the characteristics of bit planes. Proven by experimental results, our coder outperforms traditional stereo coders by 0.2-1.8 dB. Meanwhile, it can provide embedded bit streams for adapting to the channel bandwidth.