Distributed Multi-View Video Coding Based on Compressive Sensing in
WMSN
1
Luo Hui,
2
Qi Meili,
3
Jiang Nan
1,
School of Information Engineering, East China Jiaotong Uni, lh_jxnc@163.com
2,
School of Information Engineering, East China Jiaotong Uni, miliqi@sina.cn
*3,
School of Information Engineering, East China Jiaotong Uni, jiangnan1018@gmail.com
Abstract
In wireless multimedia sensor networks (WMSN), many video sensors distributed in the same scene
acquire large quantities of data. However, sensor devices are constrained in terms of power, processing,
and bandwidth capacity. This paper presents a new framework of distributed video processing system
for WMSN that takes advantage of distributed compressive sensing (DCS) theory. In the encoder
analysis the residual joint sparse model. In the decoder, efficient reconstruction is implemented using
the GPSR algorithm to recover images. Side information is generated using the temporal correlation
between adjacent frames from a single view and the spatial correlation of the two nearest views.
Experimental results show that the method presented achieves 1-3dB improvement in PSNR compared
to the coding method of single-view videos.
Keywords:
Distributed Compressive Sensing; Multi-View; WMSN; Disparity Compensation
1. Introduction
In wireless multimedia sensor networks (WMSN), a number of redundantly distributed nodes
acquire data from different perspectives simultaneously, and report them to a central collection point
which reconstructs the multi-view images. Communication energy and bandwidth of WMSN are often
scarce resources, which imposes important constraints in terms of communication costs or bit rate.
Multi-view video data are much larger than single-view video data, so there is an urgent need to
identify a simple and high compression efficiency method of multi-view video coding.
The complexity of the encoder is 5 to 10 times that of the decoder in typical video compression
standards such as MPEG or H. 264. This cannot meet the WMSN node’s low-cost, low power
requirements
[1]
. The distributed video coding (DVC) theory, based on the Slepian-Wolf and Wyner-Ziv
coding, has recently raised many interesting research questions regarding the efficient representation of
information in correlated signals captured independently by multi-view sensors. The DVC encoder
complexity is low, and can reduce the amount of communication between nodes. However, the DVC
encoder is still using the Nyquist sampling theorem and a general transform coding method[2].The
calculation of the transform coding is computation-intensive. In 2006, D. Donoh
[3,4]
proposed a
compressed sensing (CS) theory, in which the sampling number is far less than the Nyquist sampling,
and the encoding side can reduce the compression load and increase the coding efficiency. At the same
time, the CS perception observation method can reduce the computation intensity of the transform
coding, while the CS is typically used for single-channel signal encoding. Distributed compressed
sensing (DCS)
[5]
provides a new method for multi-channel signals coding. In order for the decoder to
effectively reconstruct the images of multiple views, the DCS algorithm needs to evaluate the
correlation between the video images. Most DCS algorithms rely on sparse representation to
take advantage of the correlations between multi-view images. A series of studies on the joint sparse
model
[6]
and the training dictionary of sparse representation
[7]
were carried out. These methods use the
correlation between signals and reduce the number of observations of the encoding side, but the
reconstruction of the image is not ideal.
In this paper, we propose a system for coding the multi-view video in WMSN using a distributed
compressed sensing algorithm. Based on joint sparse representation, the decoder further exploits the
temporal correlation between multi-view images and the spatial correlation between the video sequence
in one sensor. In order to preserve good image quality, the system uses disparity estimation and
disparity compensation between multi-view images for more precise side information and to optimize
Distributed Multi-View Video Coding Based on Compressive Sensing in WMSN
Luo Hui, Qi Meili, Jiang Nan
Advances in information Sciences and Service Sciences(AISS)
Volume4, Number23, Dec 2012
doi: 10.4156/AISS.vol4.issue23.87