Quality Assessment of Contrast-Altered Images
Min Liu
†
, Ke Gu
§
, Guangtao Zhai
†
, Jiantao Zhou
[
, and Weisi Lin
§
†
Insti. of Image Commu. & Infor. Proce., Shanghai Jiao Tong University, China
§
School of Computer Engineering, Nanyang Technological University, Singapore
[
Department of Computer and Information Science, University of Macau
Abstract—In image / video systems, the contrast adjustment
which manages to enhance the visual quality is nowadays an
important research topic. Yet very limited efforts have been
devoted to the exploration of image quality assessment (IQA)
for contrast adjustment. To address the problem, this paper
proposes a novel reduced-reference (RR) IQA metric with the
integration of bottom-up and top-down strategies. The former
one stems from the recently revealed free energy theory which
tells that the human visual system always seeks to understand
an input image by the uncertainty removal, while the latter
one is towards using the symmetric K-L divergence to compare
the histogram of the contrast-altered image with that of the
reference image. The bottom-up and top-down strategies are
lastly combined to derive the Reduced-reference Contrast-altered
Image Quality Measure (RCIQM). A comparison with numerous
existing IQA models is conducted on contrast related CID2013,
CCID2014, CSIQ, TID2008 and TID2013 databases, and results
validate the superiority of the proposed technique.
1
Index Terms—Contrast alteration, image quality assessment
(IQA), reduced-reference (RR), hybrid parametric and non-
parametric model (HPNP), bottom-up, top-down
I. INTRODUCTION
The importance of visual media, which in most conditions
are provided to human consumers, have been realized lately.
As the users’ requirements for high-quality images / videos
are increasingly rising, a reliable system to evaluate, control
and improve the users’ quality of experience (QoE) is highly
required. This gives rise to the demand of faithful metrics of
image quality assessment (IQA) for predicting the quality in
accordance with human visual perception [1].
With respect to the accessibility of the original references,
objective IQA metrics are mostly classified into three types: 1)
full-reference (FR); 2) reduced-reference (RR); 3) no-reference
(NR). Depending on the supposition of structural variations
being extremely vital in quality perception, the last few years
have witnessed the emergence of a vast majority of FR IQA
metrics [2-6]. Under the condition of partial reference image or
several extracted features being available as side information,
RR-IQA has a broader range of practical scenarios. Guided
by the recent discovery of free energy theory [7], we lately
designed the free energy based distortion metric (FEDM) [8]
by simulating the internal generative model of human brain to
detect input visual signals. Exploiting a set of filters and valid
pooling strategy, structural degradation model (SDM) [9] has
1
This work was supported in part by NSFC (61025005, 61371146,
61221001, 61390514, 61402547) and Macau Science and Technology De-
velopment Fund under grant FDCT/046/2014/A1.
managed to modify FR SSIM into valid RR IQA techniques
with only a few numbers as RR information.
Despite the prosperity and successfulness of IQA studies,
very limited efforts have been devoted to the field of IQA with
contrast change [10]. Moreover, existing IQA algorithms do
not work validly in this field. As a matter of fact, contrast is an
important research topic [11], which has practical applications
such as contrast enhancement technologies [12-13]. This mo-
tivates the design of a novel dedicated contrast-changed image
database (CID2013) [14], including 400 contrast-changed im-
ages by mean shifting and four kinds of transfer mappings,
and its advanced version (CCID2014) [15].
In this paper we further dig into the issue of contrast-
changed IQA, and develop a new RR IQA model with the
combination of bottom-up and top-down strategies. Relative
to the frequently seen distortion types, e.g. JPEG / JPEG2000
compressions, the human visual sensation of image contrast
(mainly including brightness and contrast alteration) is more
prone to the aesthetic quality assessment [16], and therefore
inclines to the measurement in visual and psychological fields.
A recently revealed free energy principle illustrates that the
HVS always tries to perceive a visual signal by reducing the
uncertain portion and measures the psychovisual quality as
the agreement between an image and its output of the internal
generative model. With this, we evaluate the visual quality
of contrast-altered images in the bottom-up model based on
the internal generative mechanism, which is constructed by
the non-parametric autoregressive (AR) model via perceptual
information for weighting.
On the other hand, as pointed out in several existing con-
trast enhancement methods [12], the histogram modification
can result in the contrast adjustment and largely influence
users’ experiences. The top-down strategy aims to compare
two distances between histograms; one is of the contrast-
adjusted image and its original counterpart, and the other is
of the contrast-altered image and the one created from the
original image through histogram equalization. The Kullback-
Leibler (K-L) divergence, one of the most popular information-
theoretic “distances” comparing two probability distributions,
is naturally taken into account. But the K-L divergence is
non-symmetric and brings unstable results in calculation. So
we use the symmetrized and smoothed Jensen-Shannon (JS)
divergence [17] to compute the two distances stated above.
Finally, the bottom-up and top-down strategies are combined
to develop the Reduced-reference Contrast-changed Image
Quality Measure (RCIQM), whose superiority is verified over
978-1-4799-5341-7/16/$31.00 ©2016 IEEE
2214