Information Technology Reference
In-Depth Information
Chapter 1
Automatic Prediction of Perceptual Video
Quality: Recent Trends and Research Directions
Anush K. Moorthy and Alan C. Bovik
Abstract.
Objective video quality assessment (VQA) refers to evaluation of the
quality of a video by an algorithm. The performance of any such VQA algorithm
is gaged by how well the algorithmic scores correlate with human perception of
quality. Research in the area of VQA has produced a host of full-reference (FR)
VQA algorithms. FR VQA algorithms are those in which the algorithm has access
to both the original reference video and the distorted video whose quality is being
assessed. However, in many cases, the presence of the original reference video is
not guaranteed. Hence, even though many FR VQA algorithms have been shown to
correlate well with human perception of quality, their utility remains constrained.
In this chapter, we analyze recently proposed reduced/no-reference (RR/NR) VQA
algorithms. RR VQA algorithms are those in which some information about the ref-
erence video and/or the distorting medium is embedded in the video under test. NR
VQA algorithms are expected to assess the quality of videos without any knowledge
of the reference video or the distorting medium. The utility of RR/NR algorithms
has prompted the Video Quality Experts Group (VQEG) to devote resources to-
wards forming a RR/NR test group. In this chapter, we begin by discussing how
performance of any VQA algorithm is evaluated. We introduce the popular VQEG
Phase-I VQA dataset and comment on its drawbacks. New datasets which allow for
objective evaluation of algorithms are then introduced. We then summarize some
properties of the human visual system (HVS) that are frequently utilized in devel-
oping VQA algorithms. Further, we enumerate the paths that current RR/NR VQA
algorithms take in order to evaluate visual quality. We enlist some considerations
that VQA algorithms need to consider for HD videos. We then describe exemplar
algorithms and elaborate on possible shortcomings of these algorithms. Finally, we
suggest possible future research directions in the field of VQA and conclude this
chapter.
