25-10-2012, 02:05 PM
Blind Image Quality Assessment: A Natural Scene Statistics Approach in the DCT Domain
[attachment=37089]
Abstract—
We develop an efficient general-purpose blind/
no-reference image quality assessment (IQA) algorithm using a
natural scene statistics (NSS) model of discrete cosine transform
(DCT) coefficients. The algorithm is computationally appealing,
given the availability of platforms optimized for DCT computation.
The approach relies on a simple Bayesian inference model
to predict image quality scores given certain extracted features.
The features are based on an NSS model of the image DCT
coefficients. The estimated parameters of the model are utilized
to form features that are indicative of perceptual quality. These
features are used in a simple Bayesian inference approach to
predict quality scores. The resulting algorithm, which we name
BLIINDS-II, requires minimal training and adopts a simple
probabilistic model for score prediction. Given the extracted
features from a test image, the quality score that maximizes
the probability of the empirically determined inference model
is chosen as the predicted quality score of that image. When
tested on the LIVE IQA database, BLIINDS-II is shown to
correlate highly with human judgments of quality, at a level
that is competitive with the popular SSIM index.
INTRODUCTION
THE UBIQUITY of transmitted digital visual information
in daily and professional life, and the broad range of
applications that rely on it, such as personal digital assistants,
high-definition televisions, internet video streaming, and video
on demand, necessitate the means to evaluate the visual quality
of this information. The various stages of the pipeline through
which an image passes can introduce distortions to the image,
beginning with its capture until its consumption by a viewer.
The acquisition, digitization, compression, storage, transmission,
and display processes all introduce modifications to the
original image. These modifications, also termed distortions or
impairments, may or may not be perceptually visible to human
viewers. If visible, they exhibit varying levels of annoyance.
OVERVIEW OF THE METHOD
We will refer to undistorted images captured by imaging
devices that sense radiation from the visible spectrum as
natural scenes, and statistical models built for undistorted
natural scenes as NSS models. Deviations from NSS models,
caused by the introduction of distortions to images, can be
used to predict the perceptual quality of the image. The
model-based NSS-IQA approach developed here is a process
of feature extraction from the image, followed by statistical
modeling of the extracted features. Purely NSS-based IQA
approaches require the development of a distance measure
between a given distorted test image and the NSS model.
This leads to the question of what constitutes appropriate
and perceptually meaningful distance measures between
distorted image features and NSS models. The Kullback–
Leibler divergence [21] as well as other distance measures
have been used for this purpose, but no perceptual justification
has been provided for its use.
MODEL-BASED DCT DOMAIN NSS FEATURES
We propose a parametric model to model the extracted
local DCT coefficients. The parameters of the model are
then utilized to extract features for perceptual quality score
prediction.We extract a small number of model-based features
(only four), as described next. Additionally, toward the end of
this section we point out the challenge of blindly predicting
visual quality across multiple distortions types, and we explain
the importance of multiscale feature extraction.