Multiple Description Coding Method Based On Mixed-norm Algorithm Compressive Sensing

When signal in error-prone channel, transmitted packets often generate data errors or data loss and other issues under the influence of noise. It makes difficulties for re-ceiving end to decode effectively and recover the signal. Multiple Description Coding (MDC) provides a reliable method to solve data real-time transmission under unreliable channel and receives more and more attention. The original signal is divided into dif-ferent descriptions according to certain rules through MDC and the descriptions are transmitted by independent channels. Receiving end only needs one complete descrip-tion and it can reconstruct the signal whose quality can be received. Through this kind of method, the ability of anti-loss of the transmission system is improved greatly and the robustness of the signal is enhanced.The quality of reconstructed signal through MDC is dependent on the number of received descriptions. The more descriptions received, the better the recovery quality is. But with the number of descriptions increase, the complexity of the encoding end rises while the efficiency of the encoding end reduces. Compressive sensing theory which rised in recent years provides a new method to solve this problem.Compressive sensing theory obtains the whole information of signal with the sam-ple number much less than Nyquist Sampling Theorem and each measurement can be regarded as a description of the original signal. Combining the Compressive sensing theory with MDC provides a new idea for the reliable transmission and efficient coding under error-prone channel.This paper focuses on the two-dimension image signal MDC based on Compres-sive sensing. Obtain the high and low frequency sub-band of image through wavelet transformation. Because the high frequency sub-band coefficients are more sparse, we use Compressive sensing theory to measure them. Divide all the code into two descrip-tions with quincunx downsampling method and transmit them through separated and independent channels. At the receiving end, use mixed-norm algorithm, BP algorithm and OMP algorithm to reconstruct the image.Compared with the BP algorithm based on single norm and the OMP algorithm based on greedy iteration, mixed-norm algorithm requires fewer conditions than them. Algorithm forms the limitation of noise distribution by itself, no longer needs the noise distribution and the sparsity of the signal. The ability of adapting the noise improves greatly.Experimental results show that compared with the single description, MDC can improve the anti-loss ability of signal effectively. In the noise system, mixed-norm al-gorithm which has a good anti-noise performance can obtain higher reconstructed qual-ity than BP algorithm and OMP algorithm and it is more suitable for the real communi-cation environment.