Article
Salient Region Guided Blind Image Sharpness Assessment
Siqi Liu
1,2,†
, Shaode Yu
1,2,†
, Yanming Zhao
1,2
, Zhulin Tao
1,2
, Hang Yu
3
and Libiao Jin
1,2,
*
Citation: Liu, S.; Yu, S.; Zhao, Y.; Tao,
Z.; Yu, H.; Jin, L. Salient region
guided image sharpness assessment.
Sensors 2021, 21, 3963. https://
doi.org/10.3390/s21123963
Academic Editors: Nikolaos Thomos
and Eirina Bourtsoulatze
Received: 4 April 2021
Accepted: 4 June 2021
Published: 8 June 2021
Publisher’s Note: MDPI stays neutral
with regard to jurisdictional claims in
published maps and institutional affil-
iations.
Copyright: © 2021 by the authors.
Licensee MDPI, Basel, Switzerland.
This article is an open access article
distributed under the terms and
conditions of the Creative Commons
Attribution (CC BY) license (https://
creativecommons.org/licenses/by/
4.0/).
1
Key Laboratory of Convergent Media and Intelligent Technology (Communication University of China),
Ministry of Education, Beijing 100024, China; liu47@cuc.edu.cn (S.L.); yushaodecuc@cuc.edu.cn (S.Y.);
yanmingzhao@cuc.edu.cn (Y.Z.); taozl@cuc.edu.cn (Z.T.)
2
School of Information and Communication Engineering, Communication University of China,
Beijing 100024, China
3
School of Aerospace Science and Technology, Xidian University, Xi’an 710126, China; hyu@xidian.edu.cn
* Correspondence: libiao@cuc.edu.cn
† These authors contributed equally to this work.
Abstract:
Salient regions provide important cues for scene understanding to the human vision system.
However, whether the detected salient regions are helpful in image blur estimation is unknown. In
this study, a salient region guided blind image sharpness assessment (BISA) framework is proposed,
and the effect of the detected salient regions on the BISA performance is investigated. Specifically,
three salient region detection (SRD) methods and ten BISA models are jointly explored, during
which the output saliency maps from SRD methods are re-organized as the input of BISA models.
Consequently, the change in BISA metric values can be quantified and then directly related to the
difference in BISA model inputs. Finally, experiments are conducted on three Gaussian blurring
image databases, and the BISA prediction performance is evaluated. The comparison results indicate
that salient region input can help achieve a close and sometimes superior performance to a BISA
model over the whole image input. When using the center region input as the baseline, the detected
salient regions from the saliency optimization from robust background detection (SORBD) method
lead to consistently better score prediction, regardless of the BISA model. Based on the proposed
hybrid framework, this study reveals that saliency detection benefits image blur estimation, while
how to properly incorporate SRD methods and BISA models to improve the score prediction will be
explored in our future work.
Keywords:
saliency detection; image sharpness; image quality; Gaussian blurring; human
vision system
1. Introduction
Human vision system (HVS) is verified as sensitive to the most conspicuous regions
in a visual scene, and selective attention is paid to those regions of interest [
1
,
2
]. At
each moment, the human brain needs to tackle massive messages. Since the amount of
information sensed is too high to be completely processed, the brain prioritizes salient
regions as the most important cues for follow-up analysis [
1
]. Emperical and computational
studies have reported evidence of a saliency map formed in cortical brain areas or before
the primary visual cortex [
3
–
5
], and this map is used to guide human visual attention to the
most relevant regions. This finding inspires increasing applications in object segmentation
and pattern analysis for scene understanding [6–8].
To identify the most informative and useful regions or objects in an image, salient
region detection (SRD) is becoming an increasingly hot topic in the field of computer vision,
and many SRD methods have been proposed in the last two decades [
9
–
11
]. According to
the backbone techniques, these SRD methods could be grouped into classic methods and
deep-learning-based methods [
11
], and the former can be further divided into intrinsic-
and extrinsic-cue-based methods in terms of the exploited attributes. Regarding the
intrinsic cues, an input image is explored to highlight the potential target regions through
Sensors 2021, 21, 3963. https://doi.org/10.3390/s21123963 https://www.mdpi.com/journal/sensors
