Dw32759763

Mrs. Poonam Dabas, Priya Soni / International Journal of Engineering Research and
Applications (IJERA) ISSN: 2248-9622 www.ijera.com
Vol. 3, Issue 2, March -April 2013, pp.759-763
An algorithm for measurement of quality of image
Mrs. Poonam Dabas, Priya Soni
Deptt of Computer Engineering
University Institute of Engineering &Technology, Kurukshetra

Abstract
This paper, start by providing an more visible than others. Most perceptual image
overview of the measurement of quality of image. quality assessment approaches proposed in the
There are various objective methods which can literature attempt to weight different aspects of the
measure the quality of the image. The paper error signal according to their visibility, as
providing an idea which can measure the error determined by psychophysical measurements in
between the actual image and reference image. humans or physiological measurements in animals.
And can calculate the quality of image.

Keywords—
 Error Sensitivity Function
 Perceptual Quality
 Image Quality
 Structural Similarity
 Measurement of Quality Fig. 1 Interaction of components in measurement of
quality of image
I. INTRODUCTION
Various images are subject to variety of A. Pre-Processing
distortion when they are processed, compressed, Fig. 1 illustrates a generic image quality
stored and transmitted. In these applications we can assessment framework based on error sensitivity.
process the images. The only correct method to Most perceptual quality assessment models can be
quantify the image is through subjective evaluation. described with a similar diagram, although they
The subjective evaluation is usually time consuming, dicer in detail. The stages of the diagram are as
inconvenient and complex. For a computer based follows: Pre-processing. This stage typically
educational system to provide such attention, it must performs a variety of basic operations to eliminate
reason about the reference image and actual image. known distortions from the images being compared.
The goal of research in objective image quality First, the distorted and reference signals are properly
assessment is to develop quantitative measures that scaled and aligned. Second, the signal might be
can automatically predict perceived image quality. transformed into a colour space (e.g., [14]) that is
An objective image quality metric can play a variety more appropriate for the HVS. Third, quality
of roles in image processing applications. First, it assessment metrics may need to convert the digital
can be used to dynamically monitor and adjust pixel values stored in the computer memory into
image quality. For example, a network digital video luminance values of pixels on the display device
server can examine the quality of video being through point wise nonlinear transformations. Fourth,
transmitted in order to control and allocate streaming a low-pass filter simulating the points spread
resources. Second, it can be used too optimize function of the eye optics may be applied. Finally,
algorithms and parameter settings of image the reference and the distorted images may be
processing modified using an nonlinear point operation to
simulate light adaptation Pedagogical Model.
II. Image Quality Assessment Based on
B. CSF Filtering
Error CSF Filtering. The contrast sensitivity
An image signal whose quality is being
function (CSF) describes the sensitivity of the HVS
evaluated can be thought of as a sum of an to divergent spatial and temporal frequencies that are
undistorted reference signal and an error signal. A present in the visual stimulus. Some image quality
widely adopted assumption is that the loss of
metrics include a stage that weights the signal
perceptual quality is directly related to the visibility
according to this function (typically implemented
of the error signal. The simplest implementation of
using a linear filter that approximates the fre- system,
this concept is the MSE, which objectively
a quality metric can assist in the optimal design of
quantifies the strength of the error signal. But two
pre filtering and bit assignment algorithms at the
distorted images with the same MSE may have very
encoder and of optimal reconstruction, error
different types of errors, some of which are much
concealment and post filtering algorithms at the

759 | P a g e

decoder. Third, it can be used to benchmark image is essentially accomplished by simulating the
processing systems and algorithms. Objective image functional properties of early stages of the HVS, as
quality metrics can be classified according to the characterized by both psychophysical and
availability of an original (distortion-free) image, physiological experiments. Although this bottom-up
with which the distorted image is to be compared. approach to the problem has found nearly universal
Most existing approaches are known as full- acceptance, it is important to recognize its
reference, meaning that a complete reference image limitations. In particular, the HVS is a complex and
is assumed to be known. In many practical highly non-linear system, but most models of early
applications, however, the reference image is not vision are based on linear or quasi-linear operators
available, and a no-reference or “blind” quality that have been characterized using restricted and
assessment approach is desirable. In a third type of simplistic stimuli. Thus, error-sensitivity approaches
method, the reference image is only partially must rely on a number of strong assumptions and
available, in the form of a set of extracted features generalizations. These have been noted by many
made available as side information to help evaluate previous authors, and we provide only a brief
the quality of the distorted image. This is referred to summary here.
as reduced-reference quality assessment. This paper
focuses on full-reference image quality assessment. These systems provide problems for the
learner to solve without trying to connect those
C. Channel Decomposition problems to a real world situation and are designed
Channel Decomposition. The images are to teach abstract knowledge that can be transferred
typically separated into sub bands (commonly called to multiple problem solving situations.
“channels” in the psychophysics literature) that are
selective for spatial and temporal frequency as well Problems of measurement of quality of image
as orientation. While some quality assessment The Quality Definition Problem. The most
methods implement sophisticated channel fundamental problem with the traditional approach is
decompositions that are believed to be closely the definition of image quality. In particular, it is not
related to the neural networks. clear that error visibility should be equated with loss
of quality, as some distortions may be clearly visible
D. Error Normalization but not so objectionable. An obvious example would
The error (difference) between the be multiplication of the image intensities by a global
decomposed reference and distorted signals in each scale factor. The study in [29] also suggested that the
channel calculated and normalized according to a correlation between image fidelity and image quality
certain masking model, which takes into account the is only moderate.
fact that the presence of one image component will The Suprathreshold Problem. The
decrease the visibility of another image component psychophysical experiments that underlie many error
that is proximate in spatial or temporal location, sensitivity models are specifically designed to
spatial frequency, or orientation. The normalization estimate the threshold at which stimulus is just
mechanism weights the error signal in a channel by a barely visible. These measured threshold values are
space-varying visibility threshold [26]. The visibility then used to define visual error sensitivity measures,
threshold at each point is calculated based on the such as the CSF and various masking effects.
energy of the reference and/or distorted coefficients However, very few psychophysical studies indicate
in a neighbourhood (which may include coefficients whether such near-threshold models can be
from within a spatial neighbourhood of the same generalized to characterize perceptual distortions
channel as well as other channels) and the base- significantly larger than threshold levels, as is the
sensitivity for that channel. The normalization case in a majority of image processing situations. In
process is intended to convert the error into units of the suprathreshold range, can the relative visual
just noticeable difference. distortions between different channels be normalized
using the visibility thresholds? Recent efforts have
E .Error Pooling been made to incorporate suprathreshold
The final stage of all quality metrics must psychophysics for analysing image distortions.
combine the normalized error signals over the spatial
extent of the image, and across the different The Natural Image Complexity Problem.
channels, into a single value. For most quality Most psychophysical experiments are conducted
assessment methods, pooling takes the form of a using relatively simple patterns, such as spots, bars,
Minkowski norm. or sinusoidal gratings. For example, the CSF is
typically obtained from threshold experiments using
Limitations: global sinusoidal images. The masking phenomena
The underlying principle of the error- are usually characterized using a superposition of
sensitivity approach is that perceptual quality is best two (or perhaps a few) different patterns. But all
estimated by quantifying the visibility of errors. This such patterns are much simpler than real world

760 | P a g e

images, which can be thought of as a superposition Lena Image with Various Type of Distortion:-
of a much larger number of simple patterns. Can the
models for the interactions between a few simple
patterns generalize to evaluate interactions between
tens or hundreds of patterns? Is this limited number
of simple-stimulus experiments sufficient to build a
model that can predict the visual quality of system
which has an explanation planning component that
uses a substantial domain knowledge base to
construct answers to student queries in the domain of
electrical circuits.
These classifications are really points along
a continuum, and serve as good rules of thumb rather
than skill, and hence cognitive in nature. However,
the emphasis of this system is also knowledge based
and involves generating explanations and using
general pedagogical tactics for generating feedback. Fig. 2 Lena Image with Impulsive Salt Paper Noise
Generally, tutors that teach procedural skills use a
cognitive task analysis of expert behaviour, while
tutors that teach concepts and frameworks use a
larger knowledge base and place more emphasis on
communication to be effective during instruction.

II. NEW PHILOSHIPY
In previous algorithm we have seen that
this can calculate the quality of image. But in this
case the MSE of all the images is almost same. But
this new algorithm will calculate the quality of
image and gives the best image quality. This will
calculate the quality of image in the case of the
distortion. This will calculate the quality of image in
the noise. This algorithm the error in the quality of
image if picture is not clear.
These Results are in agreement with visual
quality of the corresponding images. Although we Fig. 3 Lena Image with Additive White Noise
cannot define some clear criterion for subjective
quality assessment, human observer can intuitively
feel when distorted image is more or less close to the
reference image. Most human observers would agree
with the rankings given by the proposed quality
measure and in that sense these results are very
reasonable.

Steps for New Model:-
1. Given the reference image f(m,n), distorted image
p(m,n), width, height and number of pixels of the
input images and viewing distance, compute images
x(m,n) and y(m,n) using the described model of
HVS.
2. Compute the average correlation coefficient as the
average value of locally computed correlation
coefficients between images x(m,n) and y(m,n).
Fig. 4 Lena Image with Blurring
3. Compute the average correlation coefficient as
the average value of locally computed correlation
coefficients between images x(m,n) and e(m,n). III. FUTURE WORK
4. Compute Image Quality as MSE of frequency In this section, we are representing the only
domain coefficients obtained after some transform the measurement of image quality and we are
(DFT, DCT etc) discussing the various algorithms that represent the
5. Find the Image Quality Measure. measurement of quality of image.

761 | P a g e

In this paper, we have summarized the traditional context of optimization. But they are not very well
approach to image quality assessment based on error matched to perceived visual quality (e.g., [1]–[9]). In
sensitivity, and have enumerated its limitations. We the last three decades, a great deal of effort has gone
have proposed the use of structural similarity as an into the development of quality assessment methods
alternative for image quality assessment; it is useful that take advantage of known characteristics of the
to apply the SSIM index locally rather than globally. human visual system (HVS). The majority of the
First, image statistical features are usually highly proposed perceptual quality assessment models have
spatially non-stationary. Second, image distortions, followed a strategy of modifying the MSE measure
which may or may not depend on the local image so that errors are penalized in accordance with their
statistics, may also be space-variant. Third, at typical visibility. Section II summarizes this type of error-
viewing distances, only a local area in the image can sensitivity approach and discusses its difficulties and
be perceived with high resolution by the human limitations. In Section III, we describe a new
observer at one time instance. The decorrelation paradigm for quality assessment, based on the
problem when one chooses to use a Minkowski hypothesis that the HVS is highly adapted for
metric for spatially pooling errors, one is implicitly extracting structural information. As a specific
assuming that errors at different locations are example, we develop a measure of structural
statistically independent. This would be true if the similarity that compares local patterns of pixel
processing prior to the pooling eliminated intensities that have been normalized for luminance
dependencies in the input signals. Empirically, and contrast. In Section IV, we compare the test
however, this is not the case for linear channel results of different quality assessment models
decomposition methods such as the wavelet against a large set of subjective ratings gathered for
transform. It has been shown that a strong a database of 344 images compressed with JPEG and
dependency exists between intra- and inter-channel JPEG2000.
wavelet coefficients of natural images. In fact, state-
of-the-art wavelet image compression techniques IV. CONCLUSIONS
achieve their success by exploiting this strong An algorithm for image quality assessment
dependency. Psychophysically, various visual has been described. The algorithm uses a simple
masking models have been used to account for the HVS model, which is used to process input images.
interactions between coefficients. Statistically, it has CSF used in this model is not fixed; it has one user-
been shown that a well-designed nonlinear gain defined parameter, which controls attenuation at high
control model, in which parameters are optimized to frequencies. This way it is possible to get better
reduce dependencies rather than for fitting data from results than in the case when CSF with fixed
masking experiments, can greatly reduce the parameters is used. This is due to the fact that HVS
dependencies of the transform coefficients, it is treats very differently high frequency components
shown that Optimal design of transformation and present in the original image than those of noise.
masking models can reduce both statistical and Two processed images are used to compute average
perceptual dependencies. It remains to be seen how correlation coefficient, which measures the degree of
much these models can improve the performance of linear relationship between two images. This way we
the current quality assessment algorithms. take into account structural similarity between two
The Cognitive Interaction Problem. It is widely images, which is ignored by MSE-based measures.
known that cognitive understanding and interactive Finally, image quality measure is computed as the
visual processing (e.g., eye movements) influence average correlation coefficient between two input
the perceived quality of images. For example, a images modified by the average correlation
human observer will give different quality scores to coefficient between original image and error image.
the same image if s/he is provided with different This way we differentiate between random and signal
instructions. Prior information regarding the image dependant distortion, which have different impact on
content, or attention and fixation, may also affect the a human observer.
evaluation of the image quality. But most image Future research in this area should focus on
quality metrics do not consider these effects, as they finding better models for brightness perception,
are difficult to quantify and not well understood. which will include characteristics of display device.
Confused, another student may be able to help Another possible improvement is creating a CSF,
without relying for assistance. which models HVS more accurately.
The simplest and most widely used full-
reference quality metric is the mean squared error REFERENCES
(MSE), computed by averaging the squared intensity [1] S. Daly, “The visible difference predictor:
differences of distorted and reference image pixels, an algorithm for assessment of image
along with the related quantity of peak signal-to- fidelity,” in Digital images and human
noise ratio (PSNR). These are appealing because Vision, A.B. Watson (ed.), MIT Press,
they are simple to calculate, have clear physical Cambridge, MA, pp. 179-206,1993.
meanings, and are mathematically convenient in the

762 | P a g e

[2] R. J. Safranek and J.D. Johnston, “A
perceptually tuned sub band image coder
with image dependent quantization and
post-quantization data compression,” in
International Conference on Acoustics,
Speech, and Signal Processing (ICASSP),
Vol.3,1989.
[3] N. Damera-Venkata, T. D. Kite, W. Geisler,
B. L. Evans and A. C. Bovik, "Image
quality assessment based on a degradation
model," IEEE Transaction on Image
Processing, Vol. 9, No. 4, pp. 630-650,
April 2000.
[4] T. N. Pappas and R. J. Safranek, "Perceptual
criteria for image quality evaluation," in
Handbook of image and video processing,
Academic Press, May 2000.
[5] T. N. Pappas, T. A. Michel and R. O. Hinds,
“Supra-threshold perceptual image coding,”
in Proc. International Conf. on Image
Processing (ICIP), Vol. I, pp. 237-240, 1996.
[6] Z. Wang, A. C. Bovik and L. Lu, "Why is
image quality assessment so difficult?," in
Proc. International Conference on Acoustics,
Speech, and Signal Processing (ICASSP),
Vol. 4, pp. 3313-3316, 2002.
[7] Z. Wang and A. C. Bovik, "Universal image
quality index," IEEE Signal Processing
Letters, Vol. 9, No. 3, pp. 81-84, March
2002.
[8] J. S. Lim: "Two-dimensional signal and
image processing," Prentice-Hall,
Englewood Cliffs, New Jersey, 1990.
[9] J.L. Mannose and D.J. Sakrison, “The
effect of a visual fidelity criterion on the
encoding of images,” IEEE Transactions on
Information Theory, Vol. 20, pp. 525-536,
July 1974.
[10] T. N. Pappas and D. L. Neuhoff, “Least-
squares model based half toning,” IEEE
Transactions on Image Processing, Vol. 8,
pp. 1102-1116, August 1999
[11] B. Girod, “What’s wrong with mean-
squared error,” in Digital Images and
Human Vision (A. B. Watson, ed.), pp.
207–220, the MIT press, 1993.
[12] P. C. Teo and D. J. Heeger, “Perceptual
image distortion,” in Vol. 2179, pp. 127–
141, 1994.
[13] A. M. Eskicioglu and P. S. Fisher, “Image
quality measures ceptual factors,” in IEEE
vol. 43, pp. 2959–2965, December 1995.
[14] M. P. Eckert and A. P. Bradley, “Perceptual
quality metrics age fidelity and image
quality,” in vol 70,pp. 177–200, Nov.
1998.

763 | P a g e

Dw32759763

More Related Content

What's hot (18)

Viewers also liked (20)

Similar to Dw32759763 (20)

Recently uploaded (20)

Dw32759763