MODULE 4
Syllabus: Image Restoration: Degradation model, Unconstraint restoration- Lagrange multiplier and
constraint restoration
Inverse filtering- removal of blur caused by uniform linear motion, Weiner filtering,
Geometric transformations-spatial transformations
1. What is meant by image Restoration
Image restoration attempts to restore images that have been degraded – Identify the degradation
process and attempt to reverse it – Similar to image enhancement, but more objective
Degraded Image Restored Image
2. What is the difference between image enhancement and image restoration?
3. Different causes of image degradation
The different causes of image degradation are
Improper opening and closing of the shutter
Atmospheric turbulence
Misfocus of lens
Relative motion between camera and object which causes motion blur
4. Types of image blur
The process by which the original image is blurred is usually very complex and often unknown.
To simplify the calculations, the degradation is often modeled as a linear function which is often referred as
point-spread function
Image blurs can be broadly classified as (i) Gauss blur, (ii) out-of-focus blur, and (iii) motion blur.
1. Gauss Blur :
Gauss blur occurs due to long-time atmosphere exposure
Gauss blur is defined by the following point-spread function:
Here, σ is called the variance of the blur
2. Out of focus blur
This blurring is produced by a defocused optical system. It distributes a single point
uniformly over a disk surrounding the point.
The point-spread function of the out-of-focus blur is given by
where r is the radius and (cx, cy) is the centre of the out-of-focus point-spread function
3. Motion blur
Motion blur is due to relative motion between the recording device and the scene.
When an object or the camera is moved during light exposure, a motion-blurred
image is produced.
The motion blur can be in the form of a translation, a rotation, a sudden change of
scale, or some combinations of these.
4. Atmospheric turbulence blur
Atmospheric turbulence is a severe limitation in remote sensing.
Although the blur introduced by atmospheric turbulence depends on a variety of
factors like temperature, wind speed, exposure time, for long-time exposures, the
point spread function can be described reasonably well by a Gaussian function
5. Image Restoration Model
Degradation process operates on a
degradation function that operates on an
input image with an additive noise term
Input image is represented by using the
notation f(x,y), noise term can be
represented as η(x,y).These two terms when
combined gives the result as g(x,y).
If we are given g(x,y), some knowledge about the degradation function H or J and some knowledge
about the additive noise teem η(x,y), the objective of restoration is to obtain an estimate 𝑓̂(𝑥, 𝑦) of
the original image
We want the estimate to be as close as possible to the original image. The more we know about h
and η , the closer 𝑓̂(𝑥, 𝑦) will be to f(x,y)
If it is a linear position invariant process, then degraded image is given in the spatial domain by
h(x,y) is spatial representation of degradation function and symbol * represents convolution. In
frequency domain we may write this equation as
G(u,v) is the Fourier Transform of g(x,y) .H(u,v) is the Fourier Transform of h(x,y) . F(u,v) is the
Fourier Transform of f(x,y) .
6. Unconstrained linear filter – INVERSE FILTERING
The inverse filter is a straightforward image-restoration method.
If we know the exact point-spread function (PSF) model in the image-degradation system
and ignore the noise effect, the degraded image can be restored using the inverse filter
approach
The simplest approach to restoration is direct inverse filtering where we complete an
estimate of the transform of the original image simply by dividing the transform of the
degraded image G(u,v) by degradation function H(u,v)
We know that
Therefore
From the above equation we observe that we cannot recover the undegraded image
exactly because N(u,v) is a random function whose Fourier transform is not known.
Proof
degraded image is given in the spatial domain by
For simplicity, the coordinates of the image are ignored so that the equation is given as
7. Drawbacks of inverse filtering
Drawbacks of Inverse Filtering The main drawback of inverse filtering is that it is not always
possible to obtain an inverse.
For an inverse to exist, the matrix should be non-singular. In case, it is difficult to obtain
inverse filtering, the choice is to use a pseudo-inverse filter.
Another main drawback of an inverse filter is that an inverse filter will not perform well in the
presence of noise. If noise is present in the image, the inverse filter will tend to amplify noise
which is undesirable. In the presence of noise, it is better to go for a Wiener filter.
8. Wiener filtering - Minimum mean Square Error (MMSE) filtering
The wiener filtering approach uses the degradation function and statistical characteristics of noise
into the restoration process.
The objective is to find an estimate 𝒇̂ of the uncorrupted image f such that the mean
square error between them is minimized.
The error measure is given by
Where E{.} is the expected value of the argument
We assume that the noise and the image are uncorrelated one or the other has zero mean.
The gray levels in the estimate are a linear function of the levels in the degraded image
Based on these assumptions, the minimum of the error function is given in the frequency domain
by the following expression
This result is known as the Wiener filter. The filter, which consists of the terms inside the
brackets, also is commonly referred to as the minimum mean square error filter or the least square
error filter.
Case (1) : If the impact of noise is not taken into consideration then Sη(u,v)=0.
1
Then G (u , v)
H (u, v)
Thus Wiener filter reduces to inverse filter if the impact of noise is not taken into consideration
Sf
Case (2) : If H(u,v)=1, and Signal to noise ratio is high i.e., is very much greater than unity then
S
G(u,v)=1
i.e., Wiener filter acts as an all pass filter
Drawbacks of Wiener Filter: The Wiener filter requires a prior knowledge of the power spectral density
of original image which is unavailable in practice
9. SNR and MSE
A number of useful measures are based on the power spectra of noise and of the undegraded image.
One of the most important is the signal-to-noise ratio, approximated using frequency domain
quantities such as
This ratio gives a measure of the level of information-bearing signal power (i.e., of the original,
undegraded image) to the level of noise power.
An image with low noise would tend to have a high SNR and, conversely, the same image with a
higher level of noise would have a lower SNR. This ratio is an important measure used in
characterizing the performance of restoration algorithms
SNR can also be expressed as follows
Where the denominator is the Mean Sqaure Error (MSE) given by
The closer f and 𝒇̂ are, the larger he SNR will be.
10. What are the advantages of a Wiener filter over an inverse filter?
A Wiener filter is better than the inverse filter in the presence of noise because a Wiener filter uses a prior
statistical knowledge of the noise field. The transfer function of the Wiener filter is chosen to minimize
the mean square error using statistical information on both image and noise fields.
11. When will Wiener filter reduce to inverse filter?
In the absence of noise, Sη(u,v)=0.
1
Then G (u , v)
H (u, v)
Thus, a Wiener filter reduces to an inverse filter in the absence of noise.
12. Constraint Least Square Filter
The effect of information loss in the degraded image can often be mitigated by constraining the
restoration.
Constraints have the effect of adding information to the restoration process.
If these constraints represent additional knowledge of the original scene to be recovered, then this
knowledge should contribute to a more faithful restoration of the image.
Constrained restoration refers to the process of obtaining a meaningful restoration by biasing the solution
toward the minimiser of some specified constraint functional.
Constrained least-square filter is a regularisation technique which adds the Lagrange multiplier, λ, to
control the balance between noise artifacts and consistency with the observed data.
The constraint least square reconstruction is given by
Here, P(k,l) is the Fourier transform of the Laplacian filter. The filter P(k,l) has a large amplitude at
high frequencies, where the noise tends to be dominant. It modifies the denominator to reduce the noise
effects at high frequencies.
Proper choice of P(k,l) and λ can minimise higher-order derivatives. The iterations have to be monitored
to prevent noise amplification and various restoration parameters modified throughout the process.
13. Comment on the choice of regularisation parameter used in constrained least square restoration?
If the regularisation parameter is large, it leads to more regularisation, and the restored image tends to have
more ringing. With smaller values of the regularisation parameter, the restored image tends to have more
amplified noise effects.
14. Geometric transformations
Geometric transformations modify the spatial arrangement of pixels in an image.
These transformations are called rubber-sheet transformations because they may be viewed as
analogous to “printing” an image on a rubber sheet, then stretching or shrinking the sheet according to
a predefined set of rules.
Geometric transformations of digital images consist of two basic operations:
1. Spatial transformation of coordinates.
2. Intensity interpolation that assigns intensity values to the spatially transformed pixels
The transformation of coordinates may be expressed as
where (x,y) are pixel coordinates in the original image and (x’,y’)are the corresponding pixel
coordinates of the transformed image.
Our interest is in so-called affine transformations, which include scaling, translation, rotation, and
shearing. The key characteristic of an affine transformation in 2-D is that it preserves points, straight
lines, and planes.
Affine transformations can be expressed in general form using a single 3 3 × matrix as follows