The field of image and video processing generally
deals with signals that are meant for human consumption, such
as images or videos over the Internet. An image or video may go
through many stages of processing before being presented to a
human observer, and each stage of processing may introduce distortions
that could reduce the quality of the final display. For example,
images and videos are acquired by camera devices that may introduce
distortions due to optics, sensor noise, color calibration, exposure
control, camera motion etc. After acquisition, the image or video
may further be processed by a compression algorithm that reduce
the bandwidth requirements for storage or transmission. Such compression
algorithms are generally designed to achieve greater savings in
bandwidth by letting certain distortions happen to the signal.
Similarly, bit errors, which occur while an image is being transmitted
over a channel or (rarely) when it is stored, also tend to introduce
distortions. Finally, the display device used to render the final
output may introduce some of its own distortion, such as low reproduction
resolution, bad calibration etc. The amount of distortion that
each of these stages could add depends mostly on economics and/or
physical limitations of the devices.
One is obviously interested in being able to measure
the quality of an image or video, and to gauge the distortion
that has been added to it during different stages. One obvious
way of determining the quality of an image or video is to solicit
opinion from human observers. After all, these signals are meant
for human consumption. However, such a method is not feasible
not only due to the shear number of images and videos that are
"out there" but also because we want to be able to embed
quality measurement techniques into the very algorithms that process
images and videos, so that their output quality may be maximized
for a given set of resources.
The goal of research in objective image
quality assessment is to develop quantitative measures that can
automatically predict perceived image quality. Generally speaking,
an objective image quality metric can play an important role in
a broad range of applications, such as image acquisition, compression,
communication, displaying, printing, restoration, enhancement,
analysis and watermarking. First, it can be used to dynamically
monitor and adjust image quality. Second, it can be used
to optimize algorithms and parameter settings of image
processing systems. Third, it can be used to benchmark
image processing systems and algorithms.
In short, objective quality measurement (as opposed
to subjective quality assessment by human observers) seeks to
determine the quality of images or videos algorithmically. The
goal of objective quality assessment (QA) research is to design
algorithms whose quality prediction is in good agreement with
subjective scores from human observers.
Image and video QA algorithms may be classified
into three broad categories:
-
Full-Reference (FR) QA methods, in which the
QA algorithm has access to a 'perfect version' of the image
or video against which it can compare a 'distorted version'.
The 'perfect version' generally comes from a high-quality
acquisition device, before it is distorted by, say, compression
artifacts and transmission errors. However, the reference
image or video generally requires much more resources than
the distorted version, and hence FR QA is generally only used
as a tool for designing image and video processing algorithms
for in-lab testing,
and cannot be deployed as an application.
- No-Reference (NR) QA methods, in which the QA algorithm
has access only to the distorted signal and must estimate the quality
of the signal without any knowledge of the 'perfect version'.
Since NR methods do not require any reference information, they
can be used in any application where a quality measurement is
required. However, the price paid for this flexibility is in
terms of the ability of the algorithm to make accurate quality
predictions, or a limited scope of the NR QA algorithm (such
as NR QA for JPEG images only etc.).
- Reduced-Reference (RR) QA methods, in which partial information
regarding the 'perfect version' is available. A side-channel
(called an RR channel) exists through which some information regarding
the reference can be made available to the QA algorithm. RR QA algorithms
use this partial reference information to judge the quality of the distorted
signal.
