Research On Quantization And Hybrid Coding In Transform Domain Distributed Video Coding

As a new paradigm of video compression, Distributed Video Coding (DVC) shiftsinter-frame correlation exploitation from encoder to the decoder. DVC system has thecharacteristic of simple encoding, complex decoding, high compression efficiency as well asgood error-resilient performance as channel coding is applied to it. Therefore, DVC iswell-suited for the applications where computing power, storage capacity and powerconsumption are limited. The quantization and coding solution in encoding end is essential tothe R-D (rate-distortion) performance of the DVC, therefore this paper mainly focuses on thequantization and coding method in DCT domain Distributed Video Coding. The main workand achievements are as follows:1. The DCT transformation exploits the spatial redundancy between the WZ frames inDVC, and the amplitudes of the DCT coefficients reflect the changesof the image content in each frequency. Considering the distribution of the DCT coefficientsas well as the cost of coding the unimportant sub-bands, we proposed a quantization methodthat based on the characteristics of the DCT coefficients for DVC (AQSC). Experimentalresults show that the proposed AQSC achieves R-D improvements up to07dB, compared tothe TDWZ scheme.2. We proposed a particular RLC and a hybrid RLC-Turbo coding scheme (HCRT) forthe AC bands with low quantization levels in DVC. The RLC mode takes into account thepositions of the non-zero symbols during RLC encoding to effectively compress thesequential zero symbols to improve coding eff iciency. By estimating the coding rates of twomodes at the decoder, mode decision is made and sent back via FBC for hybrid coding thenext WZ frame. Experimental results show that the HCRT method can improve the R-Dperformances, e.g. up to1dB in soccer, especially for the videos like soccer of high motionintensity, where the AC zero symbols appear sequentially.3. For the reason that the error rates of the less significant bitplanes is high and thecorresponding turbo decoding efficiency is not promising, we proposed a double-layer hybridcoding scheme for the AC sub-bands in DVC. The less significant bitplanes of the ACsub-bands are RLC coded, while the more significant bitplanes remain turbo coded. This coding scheme eliminates the mode decision and the feedback channel in HCRT to reduce thecomplexity of the hybrid coding algorithm in DVC. Experimental results show that theproposed method improves the R-D performance of thevideo sequence of low motionintensity but does not reach thecompression performance of HCRT.