Signal, Image and Video Processing
https://doi.org/10.1007/s11760-022-02372-3
ORIGINAL PAPER
A Gabor feature-based full reference video quality assessment model
based on spatiotemporal slice of videos
Daniel Oppong Bediako
1
· Xuanqin Mou
1
· Maxwell Suobogbiree
2
Received: 2 January 2022 / Revised: 1 September 2022 / Accepted: 19 September 2022
© The Author(s), under exclusive licence to Springer-Verlag London Ltd., part of Springer Nature 2022
Abstract
The distortion of video is under the twin impacts of spatial and temporal distortions. Some of the video quality assessment
(VQA) metrics assess the video using a combination of these two measures. The accuracy of VQA may be reduced by
separating spatial and temporal aspects. The goal of this study was to assess the spatiotemporal structure of video quality by
using several VQA databases: (1) The 2D Gabor filter was designed to extract features from both spatial and temporal data,
respectively, from video sequences. Moreover, the real and imaginary parts of the 2D Gabor response taken from the reference
video and distortion video were then compared with the degree of similarity of the local frame quality. (2) To further enhance
the quality of the assessment, we combined the spatial similarity with that of the spatiotemporal similarity and proposed the
VQA model, named spatiotemporal slice Gabor feature similarity deviation. (3) In video quality scoring, a sequential pooling
strategy was used to assemble the quality indices of frames. (4) Experimental evaluations of the video quality database show
that the proposed metric has good consistency with subjective perception and competitive with state-of-the-art full reference
video quality assessment models.
Keywords Spatiotemporal slice Gabor feature similarity deviation (STS-GFSD) · Full reference video quality assessment
(FR-VQA) methods · 2D Gabor filter
1 Introduction
Millions of videos are uploaded to social networks and com-
munication devices each day. Video algorithms must assess
the perceptual quality of videos in a manner that is consis-
tent with human judgment. On a frame-by-frame basis, VQA
algorithms may produce direct augments of image qual-
ity assessment (IQA) approaches, but they frequently result
in poor performance. During the purchase, transmission,
compression, and reproduction processes, video experiences
various distortions, and the problem of video quality turns
into a focal concern.
There are two ways to measure video quality. The first
is through subjective evaluation, which takes many subjects
B Daniel Oppong Bediako
Xuanqin Mou
xqmou@mail.xjtu.edu
1
Institute of Image Processing and Pattern Recognition, Xi’an
Jiaotong University, Xi’an, China
2
School of Advance Technologies, Engineering and Science,
ACCRA Institute of Technology, Accra, Ghana
to give mean opinion scores (MOS) or difference mean
opinion scores (DMOS) for each test video. The other is by
using objective VQA algorithms. This type of method is an
effective way to evaluate video quality and save time, costs,
and labor.
The objective method has three categories. The full
reference (FR) algorithms depend on measuring the error
difference (signal difference) between the distorted and
the reference video in such a way to imitate the human
visual error sensitivity features. Usually, such approaches
provide a reference video that is accessible for comparison.
Reduce reference (RR) algorithms are suitable for some
limited situations on bandwidth that predict video quality
by using partial reference video information. No reference
(NR) algorithm assesses the quality without referring to the
original high-quality image/video. In this paper, we address
the task of FR VQA in which a reference video is available
to the VQA algorithm.
Most VAQ algorithms are based on a frame-by-frame mea-
surement. Wang et al. [1] were the first to extend the use of
the structure similarity (SSIM) index to VQA. Subsequently,
a three-SSIM algorithm [2] was proposed that further consid-
123
Content courtesy of Springer Nature, terms of use apply. Rights reserved.