Study On The Multiple Description Coding Method Based On Scalar Quantized Compressed Sensing

With the development of the internet and wireless mobile communication, there is an increasing demand for mobile multimedia services. However, the potential packet loss in network and data error or loss in the wireless channel lead to the image quality declining in receiving end. Due to satisfying the requirement of real-time data transmission and reducing data distortion, Multiple Description Coding (MDC) method is concerned widely. The MDC method divides the original image into several independent descriptions and transmits them through different channels to the decoder. Each description can independently recover the image with acceptable quality. At present the existing problem of most MDC methods is:the resistance of packet loss or error and the quality of reconstructed signal depend on the number of descriptions received. But as the description number increases, the coding complexity becomes very high and coding efficiency will be lower. Compressed Sensing (CS) theory developed in recent years offers a new way to solve this problem.Compressed Sensing, which can use the frequency that far less than the required sampling frequency of the Nyquist Sampling Theorem to sample the signal and accurately reconstruct the signal, has attracted much public attention in the area of signal processing. The CS theory can obtain the useful information in the signal through a few measurements. And each measurement can be regarded as a description of the original signal. The reconstructed quality is only related to the number of measurements.So applying the CS theory to MDC is the main research direction of the paper.This paper first introduces the research background and the significance of the subject, pointing out the direction of the research. Secondly, it gives a brief introduction on the CS theory and the performance comparison of several reconstruction algorithm. Then the paper introduces some of the basic content of the Block Compressed Sensing (BCS) and Scalar Quantization, and analyzes the advantages and disadvantages of BCS according to the experimental results. Finally the Multiple Description Coding method Based on Compressed Sensing (CS-MDC) is introduced. The method first partitions an image into several sub-images by interleaving extraction, and makes random measurements of each sub-image with BCS, and then forms multiple descriptions after quantizing and packing. At the decoding end, the method reconstructs the original image approximately or exactly with the received bit streams by solving an optimization problem. Block strategy ensures that the complexity of measurement process does not change due to image size, so the method is suitable for handling high-resolution images. Due to the low bits utilization rate and large quantization error, we replace the uniform quantization with the scalar quantization. Experimental results show that the Multiple Description Coding Method Based on Scalar Quantized Compressed Sensing is more robust to packet loss or bit error and has higher encoding efficiency.