Game Development Reference
In-Depth Information
Chapter 11
Image and Video Quality Assessment
Recently, there is a striking rise in the interest and demand of accurate image quality
assessment algorithms. Image and video quality assessment plays an important role
in optimizing, benchmarking, monitoring, and inspecting multimedia systems. Since
the human visual system (HVS) is the ultimate receiver of the visual signals, devel-
oping quality assessment algorithms that align well with human visual perception
is crucial in system design and optimization. The straightforward way to evaluate
the quality is subjective viewing by many observers with standard procedures. How-
ever, it is time-consuming, expensive, and also difficult for online applications. In
this chapter, we will review the recent advances of objective quality assessment,
including the image, video, and 3D-video quality assessment.
11.1 Image Quality Assessment
Image Quality Assessment (IQA) can be categorized into full, reduced, and no
reference according to the availability of reference image. In the literature, most
IQA methods assume that the undistorted reference image is fully available. How-
ever, in many practical applications it is not realistic to get access to the reference
images. Therefore, building IQAmethods that depend onmuch less information from
the reference image is another research focus. These algorithms can be categorized
into reduced reference (RR) and no reference (NR) IQA. The NR IQA methods are
usually designed with the prior knowledge of the distortion process or the natural
image statistics models. Because of the absence of the reference image information,
they are usually less efficient in providing a high correlation with the subjective
quality evaluations. RR IQAs can achieve a good tradeoff between the FR and NR
algorithms, as they can predict the image quality in terms of a few extracted features,
which are extracted and transmitted from the sender to the receiver side. We will first
introduce the image quality assessment databases, and then introduce the FR, RR
and NR algorithms, respectively.
 
Search WWH ::




Custom Search