First International Workshop on Multimedia Analysis and Processing
(IMAP 2007)
August 15-16, 2007
Technical Sponsored by:
IEEE Visual Signal Processing and Communications
Technical Committee
[Call for
papers] [Organizing/Program
Committee] [Technical
Program] [Invited
Speakers] [Special
Sessions] [ICCCN 2007] [ISMW 2007] [Home]
Perception
Is Reality: Visual Quality Evaluation and Beyond
Weisi Lin
Abstract: As a result of strong market, technology and
standardization drives, digital visual contents have been immersing into our
work and life at an explosive rate. Visual signal may
be acquired, synthesized, enhanced, watermarked, compressed, transmitted,
stored, reconstructed, retrieved, authenticated and presented for various applications
and services. Since the human visual system (HVS) is the ultimate receiver
and appreciator for the majority of processed images and video, a perceptual
quality criterion should be used in the system design and optimization, instead
of a traditional, mathematically defined one (e.g., MSE, SNR, PSNR or their
relative). As
a result of the evolution, the HVS develops unique characteristics, which can be turned into the advantages of system designs.
To make the machine perceive as the HVS
does can result in resource savings (e.g., bandwidth, memory space, computing
power) and performance enhancement (such as the resultant visual quality, and
new functionalities).
How to evaluate the quality of visual
signal plays a central role (explicitly or implicitly) in shaping many visual
processing algorithms and systems, as well as their implementation and
optimization. A traditional quality measure does not align with the HVS
perception because of 1) not every difference in an image noticeable; 2) not
every difference receiving same attention level. Significant research effort has been made toward modeling the HVS¡¯ picture quality evaluation during the past decade, to tackle the abovementioned two
problems of a traditional measure, and to apply the resultant models to various
applications. Some systems have begun to find their uses in industries.
In this talk, we
aim at a review, systematic analysis, sharing of research and industrial
experience, and discussion of our views for future research, in the related
fields. We will start with briefing the relevant physiological and
psychological knowledge towards the HVS¡¯ visual quality perception, and presenting
the research effort in formulating the
two basic psychophysical phenomena¡ªjust-noticeable difference (JND) and visual attention
(VA). JND and VA account for the local and global characteristics of the HVS in
appreciating an image, respectively, and their models can be combined to derive
the overall visual sensitivity. A taxonomy and overview is then given for the existing perceptual visual quality metrics (PVQMs). The applications of perceptual quality
evaluation techniques are to be presented in signal compression and
enhancement. In addition, the relevant industrial deployment will be
highlighted. The last part of the talk will give our comments of the current
progress, points of further discussion and possible future research directions.
Bio: Weisi Lin
graduated from
Since 2003, Weisi
Lin has devoted to PVQM development and perception-based visual processing.
With the topics closely related to the talk, he holds nine patents and has
published more than 70 technical publications in international refereed
journals and conferences (for JND/VA modeling, full-reference and no-reference
metrics, perceptual image/video compression, HVS-based signal enhancement,
etc.). He has also been the project leader of 5 industrial projects for various
types of PVQMs, and
maintained active long-term working relationship with the companies which are
keen in perception-based technology.