Research Of Distributed Video Coding Based On Compressed Sensing

Distributed video coding(DVC) is an efficient video coding in the field of wireless communication and has the characteristics of simple coding and strong transmission robustness. As a new proposed signal processing theory in recent years, compressed sensing(CS) breaks through the limitation of traditional Nyquist sampling theorem and makes it possible for digital devices to realize low sampling rate, low computational complexity and low energy consumption. Both DVC and CS have the same requirement of encoder with low complexity. Therefore, the organic combination of the two technologies becomes a research focus. This paper does research into the distributed video coding based on compressed sensing.The main research and technical innovation includes the following aspects:(1) Starting from the analysis of the rate-distortion(RD) performance of quantization, this paper analyzes the distribution of CS measurements and proposes a quantization method by discarding part CS measurements whose absolute value are large and then quantizing the rest CS measurements.In this way the distortion is reduced and the quantization performance is improved.(2) There exists high correlation from one block to the next in the measurement domain for block-based compressed sensing of images. According to this feature, this paper employs differential pulse-code modulation(DPCM) to reduce the redundancy between two blocks and analyzes the distribution of prediction errors in DPCM system, after that DPCM is coupled with nonuniform scalar quantization to provide quantized block-based compressed sensing of images.(3) Side information(SI) generation algorithm is an important research content in DVC systems and affects the RD performance of the whole system. Conducting in-depth research into SI generation algorithms, this paper adopts the idea of auxiliary information and puts forward a SI generation algorithm with residual information. The algorithm computes the temporal correlation between pixel blocks and then sends the corresponding residual block information to the decoder for SI estimation. A mode decision algorithm which further makes the best of video movement characteristic is added into the decoding process, consequently, the estimation precision of SI is improved.