
AN ACCURATE DEEP CONVOLUTIONAL NEURAL NETWORKS MODEL FOR
NO-REFERENCE IMAGE QUALITY ASSESSMENT
Bahetiyaer Bare, Ke Li, Bo Yan
School of Computer Science, Shanghai Key Laboratory of Intelligent Information Processing,
Fudan University,Shanghai 201203, China
byan@fudan.edu.cn
ABSTRACT
The goal of image quality assessment (IQA) is to use com-
putational models to measure the consistency between im-
age quality and subjective evaluations. In recent years, con-
volutional neural networks (CNNs) have been widely used
in image processing community and have achieved perfor-
mance leaps than non CNNs-based methods. In this work,
we describe an accurate deep CNNs model for no-reference
IQA. Taking image patches as input, our deep CNNs model
achieves an end-to-end method without any handcrafted fea-
tures and pre-processing procedures that are employed by pre-
vious no-reference IQA methods. The proposed model con-
sists of six convolutional layers, two fully connected layers,
one max pooling layer and two sum layers. The experimental
results verify that our model outperforms the state-of-the-art
no-reference IQA methods and most of the full-reference IQA
metrics.
Index Terms— Image quality assessment, Deep learning,
Convolutional neural networks
1. INTRODUCTION
With the development of social networks and the increasing
number of imaging devices, an enormous amount of visu-
al data is making its way to consumers. Digital images are
subject to a wide variety of distortions during acquisition,
processing, compression, storage, transmission and reproduc-
tion, any of which may result in a degradation of visual qual-
ity. Thus a perceptual evaluation process is needed for digital
images. While human subjective judgments of images are the
most reliable assessment, these are time consuming and diffi-
cult to obtain. Thus image quality assessment (IQA) methods
are used to automatically predict the visual quality of images.
IQA methods can be used to optimize image processing algo-
rithms and can be used to benchmark image processing sys-
tems and algorithms. Therefore, IQA plays a very important
role in image processing community.
This work is supported in part by the National Key Research and De-
velopment Plan (Grant No. 2016YFC0801005), and NSFC (Grant No.:
61522202; 61370158).
Objective quality assessment methods can be categorized
into three groups based on whether and how reference im-
ages are used. These three groups are : full-reference IQA
methods, reduced-reference IQA methods, and no-reference
IQA methods. Taking full information of original image as
reference, full-reference IQA methods perform better than
other type of methods. Among various various kind of full-
reference IQA methods, structural similarity [1] (SSIM) in-
dex is a new standard for image processing applications. It
has better quality prediction results than classic methods like
PSNR or MSE due to taking structural similarity as an im-
portant factor. In [2], Zhang et al. proposed an feature simi-
larity (FSIM) index based on the fact that human visual sys-
tem (HVS) understands an image mainly according to its low-
level features. According to the experiment results, FSIM
achieves the state-of-the-art performances on various image
quality databases. It predicts image quality very similar to the
HVS.
Although full-reference IQA provides a useful and effec-
tive way to evaluate quality differences, in many applications
the reference image is not available. So, no-reference IQA is
required. Because humans often can not judge an distorted
image without reference image, it is very challenging from
a computational perspective. No-reference measures can di-
rectly quantify image degradations by exploiting features that
are discriminant for image degradations. Most successful ap-
proaches are Natural Scene Statistics (NSS) based methods.
Early NSS based methods extracted features in transforma-
tion domains via DCT transform [3] or wavelet transform [4].
However, it is very slow to extract features from transforma-
tion domain. BRISQUE [5] extracted features from the spatial
domain, which leads to a significant reduction in computation
time. Very different from NSS based methods, CORNIA [6]
demonstrates that it is possible to learn discriminant image
features directly from the raw image pixels, instead of using
handcrafted features.
In recent years, with the explosion of CNNs based meth-
ods in computer vision and image processing tasks, some
CNNs based no-reference IQA methods are emerged. A-
mong them, Kang et al. [7] proposed a CNNs model for
978-1-5090-6067-2/17/$31.00
c
2017 IEEE
Proceedings of the IEEE International Conference on Multimedia and Expo (ICME) 2017 10-14 July 2017
978-1-5090-6067-2/17/$31.00 ©2017 IEEE ICME 2017
1356