Game Development Reference
In-Depth Information
assessment algorithms. Yet, due to the complexity issues, only few of them can be
applied in optimizing the video coding performance. In this chapter, we will review
the recent advances in quality assessment algorithms from the video compression
point of view, and discuss how they are employed to optimize the perceptual quality
of the video coding system.
12.2.1 Quality Assessment
The ultimate goal of the objective quality assessment research is to design computa-
tional models that can automatically predict the perceived visual quality. According
to the availability of the “reference image”, which is considered to be distortion-free
or perfect quality, the quality assessment algorithms are classified into three types:
full, reduced, and no reference. Full reference quality assessment algorithms can be
effectively employed to evaluate and optimize the rate distortion performance of the
video coding systems, where both the original and distorted videos can be accessed.
Reduced reference quality assessment is a hot research topic originated from video
communication application, where only a set of subtracted features are transmitted to
the receiver side to assist the quality assessment. No reference quality assessment can
also serve as a quality diagnosis solution in the multimedia system, such as video
acquisition and transcoding. However, although HVS can effectively and reliably
assess the quality of without the reference image, computational models design for
no reference quality assessment is still a very difficult task. In this chapter, we will
focus on discussing the full reference objective quality assessment algorithms, which
have been extensively studied and some of them have been applied in improving the
video compression performance.
From the visible aspect of visual artifacts, the full reference quality assessment
algorithm can also be classified into near- and suprathreshold distortion models. In
the literature, both of them have been successfully applied in developing perceptual
video coding algorithms. As the pixel level distortionmay not be noticeable due to the
masking effect, near-threshold distortion models account for the threshold at which a
stimulus is just barely visible, which is also called Just Noticeable Distortion (JND).
In this way, distortion below the JND threshold cannot be perceived by HVS. By
contrast, suprathresholdmodels aim to characterize the distortions that are significant
larger than the threshold levels, so that the perceivable distortion of visual artifacts
can be accurately measured.
Generally, JND can be computed in spatial domain or transform domain (Wu
et al.
2013
). In pixel domain, the psychophysical HVS features such as luminance
adaptation and texture masking effects are taken into account (Yang et al.
2005
),
which can be formulated as follows:
JND
p
(
n
)
=
P
L
(
n
)
+
P
˄
(
n
)
−
ʺ
·
min
{
P
L
(
n
),
P
˄
(
n
)
}
(12.1)
