A comparative study on video coding techniques with compressive sensing

Abstract

Compressive sensing method was proved to be able to perform lower sampling rate than the Nyquist rate yet maintaining good reconstruction quality. In this paper, we present the utilization of compressive sampling to encode video signal efficiently and compare the results with existing video coding standard, i.e. MPEG-4. A frame is first divided into blocks of identical size, and then a sparsity transform is employed to represent the block in a sparser domain. After that, the coefficients will undergo projection transform in order to reduce the data size according to a preset measurement rate. For a more prominent compression ratio, we can apply lower measurement rate. However, this rate also has to comply with the sparsity level of the signal. At the decoder side, a reconstruction algorithm will be conducted by means of basis pursuit or L1 minimization to guarantee acceptable accuracy. A greater compression factor can be achieved by integrating the motion compensation and estimation techniques with compressive sensing. Inter-frame coding will decrease the number of significant coefficients, hence enhancing the sparse property. For slow motion video, we need fewer reference frames. The group-of-picture size can be made adaptive, i.e. depending on current error level reported by a feedback link from decoder. Principally, the main difference between compressive video sensing and existing video coding is the use of projection matrix to sample the coefficients randomly. After sparsification and projection, the signal may experience subsequent processes, such as quantization, run-length coding, as well as entropy coding.