Chinese Journal of Electronics
Vol.22, No.3, July 2013
Distributed Compressive Video Sensing with
Adaptive Measurements Based on Structural
Similarity
∗
LIU Zhuo
1
, WANG Anhong
1
,ZENGBing
2
, ZHANG Xue
1
, BAI Huihui
3
and LI Zhihong
1
(1.Institute of Digital Media and Communication, Taiyuan University of Science and Technology, Taiyuan 030024, China)
(2.Hong Kong University of Science and Technology, Hong Kong SAR, China)
(3.Institute of Information Science, Beijing Jiaotong University, Beijing 100044, China)
Abstract — This paper presents a Distributed compres-
sive video sensing scheme with Adaptive measurements
(DCVS-AM). In this approach, the key frame in each
Group of pictures (GOP) is coded by Compressive sensing
(CS) with a fixed measurement rate; whereas other frames
in the same GOP are compressed by an adaptive random
projection in two stages, yielding the Adaptive compres-
sive sensing (ACS) frames. The first stage uses a small
and fixed measurement rate and recovers a coarse version.
In the second stage, each coarse-version ACS-frame to-
gether with its proceeding and following key frames will go
through a joint analysis at the decoder side and the analy-
sis result – Structural similarity (SSIM) that is based on a
motion-guided interpolation and calculated in a multilevel
discrete wavelet transform domain – is sent bac k to the en-
coder side to facilitate a re-sampling of the ACS-frame with
an adaptive measurement rate. Experimental results show
that our prop osed DCVS-AM consistently outperforms the
state-of-the-art DCVS with a fixed measurement.
Key words — Distributed compressive video sensing
(DCVS), Adaptive measurement rate, Structural similar-
ity (SSIM), Discrete wavelet transform (DWT).
I. Introduction
One well-known feature of conventional video coding sys-
tems, such as MPEG and H.26x, is that they are highly asym-
metric, i.e., the encoder can be 5–10 times more complex
than the decoder. In practice, such an asymmetric topology
is very suitable to broadcasting and streaming applications
where each source video is compressed only once (at the server
side) but decoded many times (at the user side). In recent
years, however, an increasing demand for the dual scenario
(i.e., the encoding is significantly less complex than the de-
coding) has emerged in up-link communications of low-power
video capturing (via mobile cameras, wireless sensor network,
etc.), where the computing power at the video-capturing end
is highly limited.
Distributed video coding (DVC)
[1]
, built on the Splepian-
Wolf and Wyner-Ziv distributed source coding theories
[2−4]
,is
a framework developed to encode the highly-correlated video
frames independently but decode them jointly. This frame-
work has successfully shifted the computationally intensive op-
erations (such as motion estimation/compensation and intra-
prediction) to the decoder side, thus offering a good solution to
the aforementioned scenario. In this paper, we follow the idea
presented recently in Refs.[5, 6] to implement DVC through
the Compressive sensing (CS) theory
[7−9]
, leading to the Dis-
tributed compressive video sensing (DCSV) framework. In
this framework, each source video frame is compressed inde-
pendently by a number of random sampling operations (each
being a simple and random linear projection) so as to keep
the simplicity at the encoder side. On the other hand, mo-
tion analysis will be conducted at the decoder side, leading
to a joint and more complicated decoding to deliver a higher
performance.
Compared with the existing works in Refs.[5, 6], our con-
tributions in this paper are summarized as follows.
• The existing DCVS schemes employ a fixed measure-
ment (or sampling) rate to all frames, which ignores varia-
tions in the temporal correlation in a video sequence. In this
paper, we propose a DCVS scheme with Adaptive measure-
ments (DCVS-AM) over different frames so as to produce a
better coding performance.
• The actual measurement rate for each frame is deter-
mined in our paper according to the popular Structural simi-
larity (SSIM) metric – an objective assessment of image quality
∗
Manuscript Received July 2012; Accepted Sept. 2012. This work is supported in part by Sino-Singapore Joint Research Projet
(No.2010DFA11010), the National Natural Science Foundation of China (No.61073142, No.61272262, No.61210006, No.61272051), the
Doctor Startup Foundation of TYUST (No.20092011), the International Cooperative Program of Shanxi Province (No.2011081055), the
Shanxi Provincial Foundation for Leaders of Disciplines in Science (No.20111022), Shanxi Province Talent Introduction and Development
Fund (2011), Shanxi Provincial Natural Science Foundation (No.2012011014-3), the College Students Innovative Program of Taiyuan
(No.120164077).