i rimage restoration chapter 5 prof. vidya manian depppggt ...manian/chapter5ip.pdfconstrained least...
TRANSCRIPT
![Page 1: I RImage Restoration Chapter 5 Prof. Vidya Manian Depppggt ...manian/chapter5IP.pdfConstrained Least Sqqguares Filtering With Wiener filtering we have a K as a constant of the ratio](https://reader033.vdocuments.us/reader033/viewer/2022041702/5e4286345a1b89656600351a/html5/thumbnails/1.jpg)
Image ProcessingImage Processing
I RImage RestorationChapter 5
Prof. Vidya ManianDept. of Electrical and Computer Engineeringp p g g
INEL 5327 ECE, UPRM1
![Page 2: I RImage Restoration Chapter 5 Prof. Vidya Manian Depppggt ...manian/chapter5IP.pdfConstrained Least Sqqguares Filtering With Wiener filtering we have a K as a constant of the ratio](https://reader033.vdocuments.us/reader033/viewer/2022041702/5e4286345a1b89656600351a/html5/thumbnails/2.jpg)
Overview
A model of the Image Degradation/Restoration Process/ Noise Models Restoration in the Presence of Noise Only-Spatial
FilteringFiltering Periodic Noise Reduction by Frequency Domain Filtering Estimating the Degradation Functionsg g Inverse Filtering Minimum Mean Square error(Wiener) Filtering Constrained Least Squares Filtering Geometric Mean Filter Geometric Transformations Geometric Transformations
INEL 5327 ECE, UPRM2
![Page 3: I RImage Restoration Chapter 5 Prof. Vidya Manian Depppggt ...manian/chapter5IP.pdfConstrained Least Sqqguares Filtering With Wiener filtering we have a K as a constant of the ratio](https://reader033.vdocuments.us/reader033/viewer/2022041702/5e4286345a1b89656600351a/html5/thumbnails/3.jpg)
A model of the Image D ti /R t ti PDegration/Restoration Process
f(x,y) : image function
H : degraded functionH : degraded function
g(x,y): degraded image
η(x,y): additive noise term
INEL 5327 ECE, UPRM3
![Page 4: I RImage Restoration Chapter 5 Prof. Vidya Manian Depppggt ...manian/chapter5IP.pdfConstrained Least Sqqguares Filtering With Wiener filtering we have a K as a constant of the ratio](https://reader033.vdocuments.us/reader033/viewer/2022041702/5e4286345a1b89656600351a/html5/thumbnails/4.jpg)
A model of the Image D ti /R t ti PDegration/Restoration Process
Is H is a linear, position-invariant process, then the degraded image is given in the spatial domain
g(x,y) = h(x,y) * f(x,y) + η(x,y) (5.1-1)
Frequency domain representation
G( ) H( )F( ) + N( ) (5 1 2)G(u,v) = H(u,v)F(u,v) + N(u,v) (5.1-2)
INEL 5327 ECE, UPRM4
![Page 5: I RImage Restoration Chapter 5 Prof. Vidya Manian Depppggt ...manian/chapter5IP.pdfConstrained Least Sqqguares Filtering With Wiener filtering we have a K as a constant of the ratio](https://reader033.vdocuments.us/reader033/viewer/2022041702/5e4286345a1b89656600351a/html5/thumbnails/5.jpg)
Noise Models - Gaussian Noise
Gaussian Noise Gaussian Noise
(5.2-1)
z: gray level
μ: mean
σ: standard deviation
σ2: variance
INEL 5327 ECE, UPRM5
![Page 6: I RImage Restoration Chapter 5 Prof. Vidya Manian Depppggt ...manian/chapter5IP.pdfConstrained Least Sqqguares Filtering With Wiener filtering we have a K as a constant of the ratio](https://reader033.vdocuments.us/reader033/viewer/2022041702/5e4286345a1b89656600351a/html5/thumbnails/6.jpg)
Noise Models - Gaussian Noise
Gaussian Noise Gaussian Noise
INEL 5327 ECE, UPRM6
![Page 7: I RImage Restoration Chapter 5 Prof. Vidya Manian Depppggt ...manian/chapter5IP.pdfConstrained Least Sqqguares Filtering With Wiener filtering we have a K as a constant of the ratio](https://reader033.vdocuments.us/reader033/viewer/2022041702/5e4286345a1b89656600351a/html5/thumbnails/7.jpg)
Noise Models - Rayleigh Noisey g
Rayleigh Noise Rayleigh Noise
(5.2-2)for z>= a
for z < a
INEL 5327 ECE, UPRM7
![Page 8: I RImage Restoration Chapter 5 Prof. Vidya Manian Depppggt ...manian/chapter5IP.pdfConstrained Least Sqqguares Filtering With Wiener filtering we have a K as a constant of the ratio](https://reader033.vdocuments.us/reader033/viewer/2022041702/5e4286345a1b89656600351a/html5/thumbnails/8.jpg)
Noise Models - Rayleigh Noisey g
Rayleigh Noise Rayleigh Noise
INEL 5327 ECE, UPRM8
![Page 9: I RImage Restoration Chapter 5 Prof. Vidya Manian Depppggt ...manian/chapter5IP.pdfConstrained Least Sqqguares Filtering With Wiener filtering we have a K as a constant of the ratio](https://reader033.vdocuments.us/reader033/viewer/2022041702/5e4286345a1b89656600351a/html5/thumbnails/9.jpg)
Noise Models – Erlang (Gamma) Noiseg ( )
Erlang Noise Erlang Noise
(5.2-5)for z>= 0
for z < 0
INEL 5327 ECE, UPRM9
![Page 10: I RImage Restoration Chapter 5 Prof. Vidya Manian Depppggt ...manian/chapter5IP.pdfConstrained Least Sqqguares Filtering With Wiener filtering we have a K as a constant of the ratio](https://reader033.vdocuments.us/reader033/viewer/2022041702/5e4286345a1b89656600351a/html5/thumbnails/10.jpg)
Noise Models - Erlang (Gamma) Noiseg ( )
Erlang Noise Erlang Noise
INEL 5327 ECE, UPRM10
![Page 11: I RImage Restoration Chapter 5 Prof. Vidya Manian Depppggt ...manian/chapter5IP.pdfConstrained Least Sqqguares Filtering With Wiener filtering we have a K as a constant of the ratio](https://reader033.vdocuments.us/reader033/viewer/2022041702/5e4286345a1b89656600351a/html5/thumbnails/11.jpg)
Noise Models – Exponential Noisep
Exponential Noise Exponential Noise
(5.2-8)for z>= 0
for z < 0
INEL 5327 ECE, UPRM11
![Page 12: I RImage Restoration Chapter 5 Prof. Vidya Manian Depppggt ...manian/chapter5IP.pdfConstrained Least Sqqguares Filtering With Wiener filtering we have a K as a constant of the ratio](https://reader033.vdocuments.us/reader033/viewer/2022041702/5e4286345a1b89656600351a/html5/thumbnails/12.jpg)
Noise Models - Exponential Noisep
Exponential Noise Exponential Noise
INEL 5327 ECE, UPRM12
![Page 13: I RImage Restoration Chapter 5 Prof. Vidya Manian Depppggt ...manian/chapter5IP.pdfConstrained Least Sqqguares Filtering With Wiener filtering we have a K as a constant of the ratio](https://reader033.vdocuments.us/reader033/viewer/2022041702/5e4286345a1b89656600351a/html5/thumbnails/13.jpg)
Noise Models – Uniform Noise
Uniform Noise Uniform Noise
(5.2-11)for a<= z <=b
otherwise
INEL 5327 ECE, UPRM13
![Page 14: I RImage Restoration Chapter 5 Prof. Vidya Manian Depppggt ...manian/chapter5IP.pdfConstrained Least Sqqguares Filtering With Wiener filtering we have a K as a constant of the ratio](https://reader033.vdocuments.us/reader033/viewer/2022041702/5e4286345a1b89656600351a/html5/thumbnails/14.jpg)
Noise Models - Uniform Noise
Uniform Noise Uniform Noise
INEL 5327 ECE, UPRM14
![Page 15: I RImage Restoration Chapter 5 Prof. Vidya Manian Depppggt ...manian/chapter5IP.pdfConstrained Least Sqqguares Filtering With Wiener filtering we have a K as a constant of the ratio](https://reader033.vdocuments.us/reader033/viewer/2022041702/5e4286345a1b89656600351a/html5/thumbnails/15.jpg)
Noise Models – Impulse (salt-and-) N ipepper) Noise
Impulse Noise Impulse Noise
(5.2-14)for z = a
for z = b
otherwise
INEL 5327 ECE, UPRM15
![Page 16: I RImage Restoration Chapter 5 Prof. Vidya Manian Depppggt ...manian/chapter5IP.pdfConstrained Least Sqqguares Filtering With Wiener filtering we have a K as a constant of the ratio](https://reader033.vdocuments.us/reader033/viewer/2022041702/5e4286345a1b89656600351a/html5/thumbnails/16.jpg)
Noise Models - Impulse (salt-and-) N ipepper) Noise
Impulse Noise Impulse Noise
INEL 5327 ECE, UPRM16
![Page 17: I RImage Restoration Chapter 5 Prof. Vidya Manian Depppggt ...manian/chapter5IP.pdfConstrained Least Sqqguares Filtering With Wiener filtering we have a K as a constant of the ratio](https://reader033.vdocuments.us/reader033/viewer/2022041702/5e4286345a1b89656600351a/html5/thumbnails/17.jpg)
Noise Models – Periodic Noise
Periodic Noise Periodic Noise
INEL 5327 ECE, UPRM17
![Page 18: I RImage Restoration Chapter 5 Prof. Vidya Manian Depppggt ...manian/chapter5IP.pdfConstrained Least Sqqguares Filtering With Wiener filtering we have a K as a constant of the ratio](https://reader033.vdocuments.us/reader033/viewer/2022041702/5e4286345a1b89656600351a/html5/thumbnails/18.jpg)
Restoration in the Presence of Noise O l S ti l Filt iOnly-Spatial Filtering
When the only degradation present is noise. When the only degradation present is noise.
g(x,y) = f(x,y) + η(x,y) (5.3-1)g( ,y) ( ,y) η( ,y) ( )
Frequency domain representation
G(u,v) = F(u,v) + N(u,v) (5.1-2)
INEL 5327 ECE, UPRM18
![Page 19: I RImage Restoration Chapter 5 Prof. Vidya Manian Depppggt ...manian/chapter5IP.pdfConstrained Least Sqqguares Filtering With Wiener filtering we have a K as a constant of the ratio](https://reader033.vdocuments.us/reader033/viewer/2022041702/5e4286345a1b89656600351a/html5/thumbnails/19.jpg)
Mean Filters – Arithmetic mean filter
The simplest of mean filters.p
(5.3-3)
Sxy : set of coordinates in a rectangular subimageSxy : set of coordinates in a rectangular subimagewindow of size m x n, centered at point (x,y).
The value of the restored image at any point (x,y) is simply arithmetic mean computed using the pixels in the region defined by Sxy.
INEL 5327 ECE, UPRM19
![Page 20: I RImage Restoration Chapter 5 Prof. Vidya Manian Depppggt ...manian/chapter5IP.pdfConstrained Least Sqqguares Filtering With Wiener filtering we have a K as a constant of the ratio](https://reader033.vdocuments.us/reader033/viewer/2022041702/5e4286345a1b89656600351a/html5/thumbnails/20.jpg)
Mean Filters – Geometric mean filter
(5.3-4)
Each restored pixel is given by the product of the pixels in the subimage window, raised to the power 1/1/mn.
INEL 5327 ECE, UPRM20
![Page 21: I RImage Restoration Chapter 5 Prof. Vidya Manian Depppggt ...manian/chapter5IP.pdfConstrained Least Sqqguares Filtering With Wiener filtering we have a K as a constant of the ratio](https://reader033.vdocuments.us/reader033/viewer/2022041702/5e4286345a1b89656600351a/html5/thumbnails/21.jpg)
Mean Filters – Harmonic mean filter
(5.3-5)
Works well for salt noise, but fails for pepper noise.
INEL 5327 ECE, UPRM21
![Page 22: I RImage Restoration Chapter 5 Prof. Vidya Manian Depppggt ...manian/chapter5IP.pdfConstrained Least Sqqguares Filtering With Wiener filtering we have a K as a constant of the ratio](https://reader033.vdocuments.us/reader033/viewer/2022041702/5e4286345a1b89656600351a/html5/thumbnails/22.jpg)
Mean Filters – Contraharmonic mean filtfilter
(5.3-6)
Q: order of the filter.Works well for reducing or virtually eliminating the effects of salt-and-pepper noise.Q positive eliminates the pepperQ negative eliminates the salt
INEL 5327 ECE, UPRM22
![Page 23: I RImage Restoration Chapter 5 Prof. Vidya Manian Depppggt ...manian/chapter5IP.pdfConstrained Least Sqqguares Filtering With Wiener filtering we have a K as a constant of the ratio](https://reader033.vdocuments.us/reader033/viewer/2022041702/5e4286345a1b89656600351a/html5/thumbnails/23.jpg)
Mean Filters – Examples Ip
INEL 5327 ECE, UPRM23
![Page 24: I RImage Restoration Chapter 5 Prof. Vidya Manian Depppggt ...manian/chapter5IP.pdfConstrained Least Sqqguares Filtering With Wiener filtering we have a K as a constant of the ratio](https://reader033.vdocuments.us/reader033/viewer/2022041702/5e4286345a1b89656600351a/html5/thumbnails/24.jpg)
Mean Filters – Examples IIp
INEL 5327 ECE, UPRM24
![Page 25: I RImage Restoration Chapter 5 Prof. Vidya Manian Depppggt ...manian/chapter5IP.pdfConstrained Least Sqqguares Filtering With Wiener filtering we have a K as a constant of the ratio](https://reader033.vdocuments.us/reader033/viewer/2022041702/5e4286345a1b89656600351a/html5/thumbnails/25.jpg)
Order Statistics Filters
O d S i i Fil E iOrder Statistics Filters Expression
Median Filter
Max Filter
Min Filter
Mid Point Filter
Alpha-trimmed mean filter
INEL 5327 ECE, UPRM25
![Page 26: I RImage Restoration Chapter 5 Prof. Vidya Manian Depppggt ...manian/chapter5IP.pdfConstrained Least Sqqguares Filtering With Wiener filtering we have a K as a constant of the ratio](https://reader033.vdocuments.us/reader033/viewer/2022041702/5e4286345a1b89656600351a/html5/thumbnails/26.jpg)
Order Statistics Filters – Example Ip
INEL 5327 ECE, UPRM26
![Page 27: I RImage Restoration Chapter 5 Prof. Vidya Manian Depppggt ...manian/chapter5IP.pdfConstrained Least Sqqguares Filtering With Wiener filtering we have a K as a constant of the ratio](https://reader033.vdocuments.us/reader033/viewer/2022041702/5e4286345a1b89656600351a/html5/thumbnails/27.jpg)
Order Statistics Filters – Example IIp
INEL 5327 ECE, UPRM27
![Page 28: I RImage Restoration Chapter 5 Prof. Vidya Manian Depppggt ...manian/chapter5IP.pdfConstrained Least Sqqguares Filtering With Wiener filtering we have a K as a constant of the ratio](https://reader033.vdocuments.us/reader033/viewer/2022041702/5e4286345a1b89656600351a/html5/thumbnails/28.jpg)
Adaptive Filtersp
Adaptive, local noise reduction filter g(x,y) the value of the noisy image at
(x,y)
variance of the noise corrupting f(x,y) to form g(x,y)
the local mean of the pixels in Sxy
the local variance of the pixels in Sxy
INEL 5327 ECE, UPRM28
![Page 29: I RImage Restoration Chapter 5 Prof. Vidya Manian Depppggt ...manian/chapter5IP.pdfConstrained Least Sqqguares Filtering With Wiener filtering we have a K as a constant of the ratio](https://reader033.vdocuments.us/reader033/viewer/2022041702/5e4286345a1b89656600351a/html5/thumbnails/29.jpg)
Adaptive Filtersp
Adaptive median filter
Level A: A1 = Zmed – ZminA2 = Zmed ZmaxA2 = Zmed – Zmaxif A1 > 0 AND A2 < 0, Go to level BIf window size <= Smax repeat level AElse output Zmed
Le el B B1 = Z ZminLevel B: B1 = Zxy – ZminB2 = Zxy – Zmaxif B1 > 0 AND B2 < 0, output ZxyElse output Zmed
INEL 5327 ECE, UPRM29
![Page 30: I RImage Restoration Chapter 5 Prof. Vidya Manian Depppggt ...manian/chapter5IP.pdfConstrained Least Sqqguares Filtering With Wiener filtering we have a K as a constant of the ratio](https://reader033.vdocuments.us/reader033/viewer/2022041702/5e4286345a1b89656600351a/html5/thumbnails/30.jpg)
Periodic Noise Reduction
Periodic noise appears as concentrated bursts of Periodic noise appears as concentrated bursts of energy in the Fourier transform, at locations corresponding to the frequencies of the periodic p g q pinterference
Bandreject filtersj General location of the noise component(s) in the
frequency domain is approximately known Bandpass filter performs
INEL 5327 ECE, UPRM30
![Page 31: I RImage Restoration Chapter 5 Prof. Vidya Manian Depppggt ...manian/chapter5IP.pdfConstrained Least Sqqguares Filtering With Wiener filtering we have a K as a constant of the ratio](https://reader033.vdocuments.us/reader033/viewer/2022041702/5e4286345a1b89656600351a/html5/thumbnails/31.jpg)
Periodic Noise Reduction by Frequency D i Filt iDomain Filtering
Bandreject FiltersjIdeal
Butterworth bandreject
Gaussian Filters
Bandpass Filters
INEL 5327 ECE, UPRM31
![Page 32: I RImage Restoration Chapter 5 Prof. Vidya Manian Depppggt ...manian/chapter5IP.pdfConstrained Least Sqqguares Filtering With Wiener filtering we have a K as a constant of the ratio](https://reader033.vdocuments.us/reader033/viewer/2022041702/5e4286345a1b89656600351a/html5/thumbnails/32.jpg)
Periodic Noise Reduction by Frequency D i Filt i E lDomain Filtering – Example
INEL 5327 ECE, UPRM32
![Page 33: I RImage Restoration Chapter 5 Prof. Vidya Manian Depppggt ...manian/chapter5IP.pdfConstrained Least Sqqguares Filtering With Wiener filtering we have a K as a constant of the ratio](https://reader033.vdocuments.us/reader033/viewer/2022041702/5e4286345a1b89656600351a/html5/thumbnails/33.jpg)
Noise is seen as symmetric pairs of bright dots in Noise is seen as symmetric pairs of bright dots in the Fourier spectrum
Circularly symmetric bandreject filter Circularly symmetric bandreject filter Butterworth bandreject filter of order 4 with
appropriate radius and width to enclose completely appropriate radius and width to enclose completely the noise impulses
Sharp narrow filters to remove as little as possible Sharp narrow filters to remove as little as possible from the transform
INEL 5327 ECE, UPRM33
![Page 34: I RImage Restoration Chapter 5 Prof. Vidya Manian Depppggt ...manian/chapter5IP.pdfConstrained Least Sqqguares Filtering With Wiener filtering we have a K as a constant of the ratio](https://reader033.vdocuments.us/reader033/viewer/2022041702/5e4286345a1b89656600351a/html5/thumbnails/34.jpg)
Straight bandpass filtering on an image is not Straight bandpass filtering on an image is not common because it removes too much image detail
Can be done to isolate noise pattern Can be done to isolate noise pattern Notch filters Rejects or passes frequencies in predefined Rejects or passes frequencies in predefined
neighborhoods about a center frequency Should be symmetric pairs about origin Should be symmetric pairs about origin Shape of the notch can be arbitrary (eg.,
rectangular)rectangular)INEL 5327 ECE, UPRM
![Page 35: I RImage Restoration Chapter 5 Prof. Vidya Manian Depppggt ...manian/chapter5IP.pdfConstrained Least Sqqguares Filtering With Wiener filtering we have a K as a constant of the ratio](https://reader033.vdocuments.us/reader033/viewer/2022041702/5e4286345a1b89656600351a/html5/thumbnails/35.jpg)
Notch FiltersNotch Filters RejectRadius D0 with centers at (u0, v0), by simemetry at (-u0, -v0)
Ideal
Butterworth bandreject
Gaussian Filters
These three filters become highpass filters if u0 = v0 = 0
Notch Filters Pass
INEL 5327 ECE, UPRM35
![Page 36: I RImage Restoration Chapter 5 Prof. Vidya Manian Depppggt ...manian/chapter5IP.pdfConstrained Least Sqqguares Filtering With Wiener filtering we have a K as a constant of the ratio](https://reader033.vdocuments.us/reader033/viewer/2022041702/5e4286345a1b89656600351a/html5/thumbnails/36.jpg)
Figure shows constructed notch filter and noise in the
36
Figure shows constructed notch filter and noise in the spatial domain after passing through the notch filter
Optimum notch filtering Optimum notch filtering Fig 5.20 shows starlike components caused by
interference due to several sinusoidal componentsinterference due to several sinusoidal components Optimum method to minimize local variance of the
restored estimate f’(x,y)restored estimate f (x,y) Extract principal frequency component of
interference pattern N(u,v)=HNP(u,v)G(u,v)interference pattern N(u,v) HNP(u,v)G(u,v)
INEL 5327 ECE, UPRM
![Page 37: I RImage Restoration Chapter 5 Prof. Vidya Manian Depppggt ...manian/chapter5IP.pdfConstrained Least Sqqguares Filtering With Wiener filtering we have a K as a constant of the ratio](https://reader033.vdocuments.us/reader033/viewer/2022041702/5e4286345a1b89656600351a/html5/thumbnails/37.jpg)
G(uv) is FT of corrupted image
37
G(uv) is FT of corrupted image Formation of HNP(u,v) requires judgment about
what is or is not an interference spikewhat is or is not an interference spike Corresponding pattern in spatial domain after
choosing the particular filterchoosing the particular filter n(x,y)=F-1{HNP(u,v)G(u,v)}
INEL 5327 ECE, UPRM
![Page 38: I RImage Restoration Chapter 5 Prof. Vidya Manian Depppggt ...manian/chapter5IP.pdfConstrained Least Sqqguares Filtering With Wiener filtering we have a K as a constant of the ratio](https://reader033.vdocuments.us/reader033/viewer/2022041702/5e4286345a1b89656600351a/html5/thumbnails/38.jpg)
Notch Filters – Examplep
INEL 5327 ECE, UPRM38
![Page 39: I RImage Restoration Chapter 5 Prof. Vidya Manian Depppggt ...manian/chapter5IP.pdfConstrained Least Sqqguares Filtering With Wiener filtering we have a K as a constant of the ratio](https://reader033.vdocuments.us/reader033/viewer/2022041702/5e4286345a1b89656600351a/html5/thumbnails/39.jpg)
Periodic interference39
INEL 5327 ECE, UPRM
![Page 40: I RImage Restoration Chapter 5 Prof. Vidya Manian Depppggt ...manian/chapter5IP.pdfConstrained Least Sqqguares Filtering With Wiener filtering we have a K as a constant of the ratio](https://reader033.vdocuments.us/reader033/viewer/2022041702/5e4286345a1b89656600351a/html5/thumbnails/40.jpg)
Estimating the degradation functiong g By observation, experimentation and mathematical modeling Blind deconvolution: process of restoring an image by using a
40
p g g y gdegradation function that bas been estimated in some way
By observation deduce complete degradation function H(u,v) b d i f i i i ibased on assumption of position invariance
Perform experiment with the equipment used to acquire the image to estimate the degradationimage to estimate the degradation
Obtain impulse response of the degradation. A linear, space-invariant system is characterized completely by its impulse response
H(u,v)=G(u,v)/AG( ) i th FT f th b d i d A i t t G(u,v) is the FT of the observed image and A is a constant
INEL 5327 ECE, UPRM
![Page 41: I RImage Restoration Chapter 5 Prof. Vidya Manian Depppggt ...manian/chapter5IP.pdfConstrained Least Sqqguares Filtering With Wiener filtering we have a K as a constant of the ratio](https://reader033.vdocuments.us/reader033/viewer/2022041702/5e4286345a1b89656600351a/html5/thumbnails/41.jpg)
Estimation by modelingy g
Using environmental factors such as atmospheric
41
Using environmental factors such as atmospheric turbulence
INEL 5327 ECE, UPRM
![Page 42: I RImage Restoration Chapter 5 Prof. Vidya Manian Depppggt ...manian/chapter5IP.pdfConstrained Least Sqqguares Filtering With Wiener filtering we have a K as a constant of the ratio](https://reader033.vdocuments.us/reader033/viewer/2022041702/5e4286345a1b89656600351a/html5/thumbnails/42.jpg)
Estimating the Degradation Functiong g
Because the true degradation function is seldom known completely, the process of restoring ll d “bl d d l ”image is called “blind deconvolution”.
Let assume that noise is negligible, so our principal equation (5-1.1) will be:
g(x,y) = h(x,y) * f(x,y)
By Image Observation
By Experimentation
By Modeling
For example a model for atmospheric turbulence:
INEL 5327 ECE, UPRM42
![Page 43: I RImage Restoration Chapter 5 Prof. Vidya Manian Depppggt ...manian/chapter5IP.pdfConstrained Least Sqqguares Filtering With Wiener filtering we have a K as a constant of the ratio](https://reader033.vdocuments.us/reader033/viewer/2022041702/5e4286345a1b89656600351a/html5/thumbnails/43.jpg)
Estimating the Degradation Function –E lExample
INEL 5327 ECE, UPRM43
![Page 44: I RImage Restoration Chapter 5 Prof. Vidya Manian Depppggt ...manian/chapter5IP.pdfConstrained Least Sqqguares Filtering With Wiener filtering we have a K as a constant of the ratio](https://reader033.vdocuments.us/reader033/viewer/2022041702/5e4286345a1b89656600351a/html5/thumbnails/44.jpg)
Inverse Filteringg
A simple approach is direct inverse filtering to restore the image. Just dividing h f d d d f G( ) b h d d fthe transform degraded function G(u,v), by the degradation function:
If we replace G(u,v) by F(u,v)H(u,v)+ N(u,v), we obtain:
As you see, we cannot recover the undegraded image exactly because N(u,v) As you see, we cannot recover the undegraded image exactly because N(u,v) is a random function whose Fourier transform is not known. If H(u,v) is zero or very small values, then the ratio N(u,v)/H(u,v) would dominate the estimate
INEL 5327 ECE, UPRM44
![Page 45: I RImage Restoration Chapter 5 Prof. Vidya Manian Depppggt ...manian/chapter5IP.pdfConstrained Least Sqqguares Filtering With Wiener filtering we have a K as a constant of the ratio](https://reader033.vdocuments.us/reader033/viewer/2022041702/5e4286345a1b89656600351a/html5/thumbnails/45.jpg)
H(0,0) is the highest value of H(u,v)
45
H(0,0) is the highest value of H(u,v) By limiting the analysis to frequencies near origin,
the probability of encountering zeros is reducedthe probability of encountering zeros is reduced Fig. 5.27: Inverse fitlering Results are better cutting off values of the ratio Results are better cutting off values of the ratio
G(u,v)/H(u,v) outside a radius of 40, 70 and 85 resply.resply.
Cutoff is implemented by applying to the ratio a Butterworth lowpass function of order 10Butterworth lowpass function of order 10
INEL 5327 ECE, UPRM
![Page 46: I RImage Restoration Chapter 5 Prof. Vidya Manian Depppggt ...manian/chapter5IP.pdfConstrained Least Sqqguares Filtering With Wiener filtering we have a K as a constant of the ratio](https://reader033.vdocuments.us/reader033/viewer/2022041702/5e4286345a1b89656600351a/html5/thumbnails/46.jpg)
INEL 5327 ECE, UPRM
![Page 47: I RImage Restoration Chapter 5 Prof. Vidya Manian Depppggt ...manian/chapter5IP.pdfConstrained Least Sqqguares Filtering With Wiener filtering we have a K as a constant of the ratio](https://reader033.vdocuments.us/reader033/viewer/2022041702/5e4286345a1b89656600351a/html5/thumbnails/47.jpg)
Inverse Filtering – Exampleg p
INEL 5327 ECE, UPRM47
![Page 48: I RImage Restoration Chapter 5 Prof. Vidya Manian Depppggt ...manian/chapter5IP.pdfConstrained Least Sqqguares Filtering With Wiener filtering we have a K as a constant of the ratio](https://reader033.vdocuments.us/reader033/viewer/2022041702/5e4286345a1b89656600351a/html5/thumbnails/48.jpg)
Minimum Mean Square Error (Wiener) Filt iFiltering
With this filtering we have an approach to incorporate a statistical characteristics of With this filtering we have an approach to incorporate a statistical characteristics of noise into the restoration process.
degradation function = complex complex conjugate of =
= power spectrum of the noise =
t f th
If noise is zero, the power spectrum vanishes and Wiener filter reduces to the inverse
power spectrum of the undegraded image =
filter. If we consider a as a constant, we have the expression reduces to:
K is a specified constant.INEL 5327 ECE, UPRM48
![Page 49: I RImage Restoration Chapter 5 Prof. Vidya Manian Depppggt ...manian/chapter5IP.pdfConstrained Least Sqqguares Filtering With Wiener filtering we have a K as a constant of the ratio](https://reader033.vdocuments.us/reader033/viewer/2022041702/5e4286345a1b89656600351a/html5/thumbnails/49.jpg)
Wiener Filtering – Example Ig p
INEL 5327 ECE, UPRM49
![Page 50: I RImage Restoration Chapter 5 Prof. Vidya Manian Depppggt ...manian/chapter5IP.pdfConstrained Least Sqqguares Filtering With Wiener filtering we have a K as a constant of the ratio](https://reader033.vdocuments.us/reader033/viewer/2022041702/5e4286345a1b89656600351a/html5/thumbnails/50.jpg)
Wiener Filtering – Example IIg p
INEL 5327 ECE, UPRM50
![Page 51: I RImage Restoration Chapter 5 Prof. Vidya Manian Depppggt ...manian/chapter5IP.pdfConstrained Least Sqqguares Filtering With Wiener filtering we have a K as a constant of the ratio](https://reader033.vdocuments.us/reader033/viewer/2022041702/5e4286345a1b89656600351a/html5/thumbnails/51.jpg)
The restored image in the spatial domain is given
51
The restored image in the spatial domain is given by the inverse Fourier transform of the frequency domain estimate
A number of useful measures are based on the power spectra of noise and of the undegradedp p gimage: Signal –to-noise ratio, mean square error
Root mean square error Root mean square signal to noise ratio
INEL 5327 ECE, UPRM
![Page 52: I RImage Restoration Chapter 5 Prof. Vidya Manian Depppggt ...manian/chapter5IP.pdfConstrained Least Sqqguares Filtering With Wiener filtering we have a K as a constant of the ratio](https://reader033.vdocuments.us/reader033/viewer/2022041702/5e4286345a1b89656600351a/html5/thumbnails/52.jpg)
Constrained Least Squares Filteringq g
With Wiener filtering we have a K as a constant of the ratio of the power spectraWith Wiener filtering we have a K as a constant of the ratio of the power spectra.
With Constrained Least Squares Filtering is required to know the mean and variance of the noise and we can calculate these variables from a given degraded image.image.
We can express our main expression in vector-matrix form, as follows:
Subject to the constraint:
The frequency domain solution is given by the next expression:
INEL 5327 ECE, UPRM52
![Page 53: I RImage Restoration Chapter 5 Prof. Vidya Manian Depppggt ...manian/chapter5IP.pdfConstrained Least Sqqguares Filtering With Wiener filtering we have a K as a constant of the ratio](https://reader033.vdocuments.us/reader033/viewer/2022041702/5e4286345a1b89656600351a/html5/thumbnails/53.jpg)
Constrained Least Squares Filtering -E lExample
INEL 5327 ECE, UPRM53
![Page 54: I RImage Restoration Chapter 5 Prof. Vidya Manian Depppggt ...manian/chapter5IP.pdfConstrained Least Sqqguares Filtering With Wiener filtering we have a K as a constant of the ratio](https://reader033.vdocuments.us/reader033/viewer/2022041702/5e4286345a1b89656600351a/html5/thumbnails/54.jpg)
INEL 5327 ECE, UPRM54
![Page 55: I RImage Restoration Chapter 5 Prof. Vidya Manian Depppggt ...manian/chapter5IP.pdfConstrained Least Sqqguares Filtering With Wiener filtering we have a K as a constant of the ratio](https://reader033.vdocuments.us/reader033/viewer/2022041702/5e4286345a1b89656600351a/html5/thumbnails/55.jpg)
I t d ti t C t 55
Introduction to Computer Tomography –Dept. of Homeland g p y pSecurity
Supplement MaterialSupplement Material
INEL 5327 ECE, UPRM
![Page 56: I RImage Restoration Chapter 5 Prof. Vidya Manian Depppggt ...manian/chapter5IP.pdfConstrained Least Sqqguares Filtering With Wiener filtering we have a K as a constant of the ratio](https://reader033.vdocuments.us/reader033/viewer/2022041702/5e4286345a1b89656600351a/html5/thumbnails/56.jpg)
Image Reconstruction from P j tiProjections Reconstructing an image from a series of
56
projections, X-ray computed tomography (CT) Imaging a single object on a uniform background 3-D region of a human body –round object is a
tumor with higher absorption characteristics Pass a thin flat beam of X-rays from left to right,
energy absorbed more by object than background Any point in the signal is the sum of the absorption
values across the single ray in the beam d ll h ( f corresponding spatially to that point (information is
the 1-D absorption signal)INEL 5327 ECE, UPRM
![Page 57: I RImage Restoration Chapter 5 Prof. Vidya Manian Depppggt ...manian/chapter5IP.pdfConstrained Least Sqqguares Filtering With Wiener filtering we have a K as a constant of the ratio](https://reader033.vdocuments.us/reader033/viewer/2022041702/5e4286345a1b89656600351a/html5/thumbnails/57.jpg)
INEL 5327 ECE, UPRM57
![Page 58: I RImage Restoration Chapter 5 Prof. Vidya Manian Depppggt ...manian/chapter5IP.pdfConstrained Least Sqqguares Filtering With Wiener filtering we have a K as a constant of the ratio](https://reader033.vdocuments.us/reader033/viewer/2022041702/5e4286345a1b89656600351a/html5/thumbnails/58.jpg)
Backprojectionp j
Project 1-D signal back across the direction from
58
Project 1 D signal back across the direction from which the beam came
Backprojection of a 1D signal across a 2-D area is Backprojection of a 1D signal across a 2 D area is called smearing
Duplicating the image perpendicularly to the Duplicating the image perpendicularly to the direction of the beam
Create image in vertical direction Create image in vertical direction Reconstruction by adding the of backprojection Object of interestis contained in the square Object of interestis contained in the square
INEL 5327 ECE, UPRM
![Page 59: I RImage Restoration Chapter 5 Prof. Vidya Manian Depppggt ...manian/chapter5IP.pdfConstrained Least Sqqguares Filtering With Wiener filtering we have a K as a constant of the ratio](https://reader033.vdocuments.us/reader033/viewer/2022041702/5e4286345a1b89656600351a/html5/thumbnails/59.jpg)
INEL 5327 ECE, UPRM59
![Page 60: I RImage Restoration Chapter 5 Prof. Vidya Manian Depppggt ...manian/chapter5IP.pdfConstrained Least Sqqguares Filtering With Wiener filtering we have a K as a constant of the ratio](https://reader033.vdocuments.us/reader033/viewer/2022041702/5e4286345a1b89656600351a/html5/thumbnails/60.jpg)
A b f j ti i th t th f As number of projections increases , the strength of the non-intersecting backprojections decreases relative to the strength of regions in which multiple
60
relative to the strength of regions in which multiple projections intersects
As number of views increases the shape of the halo As number of views increases, the shape of the halo becomes circular
Blurring in CT reconstruction Blurring in CT reconstruction Projections 180 apart are mirror images of each
other consider angle increments halfway around a other, consider angle increments halfway around a circle to generate all projections required for reconstruction
INEL 5327 ECE, UPRM
![Page 61: I RImage Restoration Chapter 5 Prof. Vidya Manian Depppggt ...manian/chapter5IP.pdfConstrained Least Sqqguares Filtering With Wiener filtering we have a K as a constant of the ratio](https://reader033.vdocuments.us/reader033/viewer/2022041702/5e4286345a1b89656600351a/html5/thumbnails/61.jpg)
61
INEL 5327 ECE, UPRM
![Page 62: I RImage Restoration Chapter 5 Prof. Vidya Manian Depppggt ...manian/chapter5IP.pdfConstrained Least Sqqguares Filtering With Wiener filtering we have a K as a constant of the ratio](https://reader033.vdocuments.us/reader033/viewer/2022041702/5e4286345a1b89656600351a/html5/thumbnails/62.jpg)
Principles of computed t h (CT)tomography(CT) Obtain a 3D representation of the internal struction
62
Obtain a 3D representation of the internal structionof an object by X-raying the object from many different directions
X-ray beam in the form of a cone X-ray plate produces an image whose intensity at a X ray plate produces an image whose intensity at a
point is proportional to the X-ray energy impinging on that point after it has passed through the subject
2-D equivalent of backprojection Backproject the entire image and create a 3D Backproject the entire image and create a 3D
volumeINEL 5327 ECE, UPRM
![Page 63: I RImage Restoration Chapter 5 Prof. Vidya Manian Depppggt ...manian/chapter5IP.pdfConstrained Least Sqqguares Filtering With Wiener filtering we have a K as a constant of the ratio](https://reader033.vdocuments.us/reader033/viewer/2022041702/5e4286345a1b89656600351a/html5/thumbnails/63.jpg)
Repeating process through many angles and adding Repeating process through many angles and adding the backprojectiosn would result in 3D rendition of the structure of the chest cavity
63
CT gets the same information by generating slices through the body
3D representation can be obtained by stacking the slices
A CT is much more economical, number of detectors required to get a high resolution slice is much smaller than generating a complex 2D projection of the same resolution
X-ray levels and computation cost are less in the 1D projection CT approachINEL 5327 ECE, UPRM
![Page 64: I RImage Restoration Chapter 5 Prof. Vidya Manian Depppggt ...manian/chapter5IP.pdfConstrained Least Sqqguares Filtering With Wiener filtering we have a K as a constant of the ratio](https://reader033.vdocuments.us/reader033/viewer/2022041702/5e4286345a1b89656600351a/html5/thumbnails/64.jpg)
Johanna Radon-Radon transform (projecting a 2D
64
Johanna Radon Radon transform (projecting a 2D object along parallel rays
Allan M. Cormack-applied the above concept to CT Allan M. Cormack applied the above concept to CT Cormack and Hounsfield shared the 1979 Nobel
Prize in Medicine for their contributions to medical Prize in Medicine for their contributions to medical tomography
First generation (G1) CT scanners employ a ‘pencil’ First generation (G1) CT scanners employ a pencil X-ray beam and a single detector
For a given angle of ration, the source/detector For a given angle of ration, the source/detector pair is translated incrementally along the linear direction
INEL 5327 ECE, UPRM
![Page 65: I RImage Restoration Chapter 5 Prof. Vidya Manian Depppggt ...manian/chapter5IP.pdfConstrained Least Sqqguares Filtering With Wiener filtering we have a K as a constant of the ratio](https://reader033.vdocuments.us/reader033/viewer/2022041702/5e4286345a1b89656600351a/html5/thumbnails/65.jpg)
65
INEL 5327 ECE, UPRM
![Page 66: I RImage Restoration Chapter 5 Prof. Vidya Manian Depppggt ...manian/chapter5IP.pdfConstrained Least Sqqguares Filtering With Wiener filtering we have a K as a constant of the ratio](https://reader033.vdocuments.us/reader033/viewer/2022041702/5e4286345a1b89656600351a/html5/thumbnails/66.jpg)
A projection is generated by measuring the output
66
A projection is generated by measuring the output of the detector at each increment of translation
After a complete linear translation, the After a complete linear translation, the source/detector assembly is rotated and the procedure is repeated to generate another p p gprojection at a different angle. Procedure is repeated for all angles [0,180]degrees
Image is generated by backprojection. A set of cross-sectional imaages(slices) is generated
by incrementally moving the subject pst the source/detector plane INEL 5327 ECE, UPRM
![Page 67: I RImage Restoration Chapter 5 Prof. Vidya Manian Depppggt ...manian/chapter5IP.pdfConstrained Least Sqqguares Filtering With Wiener filtering we have a K as a constant of the ratio](https://reader033.vdocuments.us/reader033/viewer/2022041702/5e4286345a1b89656600351a/html5/thumbnails/67.jpg)
Stacking these images computationally produces a g g p y p3D volume of a section of the body
Second generation (G2) CT scanners –beam used is 67
g ( )in the shape of a fan
Requires fewer translations of the source/detector q /pair
Third-generation (G3) scanners –employ a bank of detectors long enough to cover the entire field of view of a wider beam
Fourth generation (G4) scanners-employ a circular ring of detection (5000) only the source has to rotate
Advantage of G3 and G4 is speed.INEL 5327 ECE, UPRM
![Page 68: I RImage Restoration Chapter 5 Prof. Vidya Manian Depppggt ...manian/chapter5IP.pdfConstrained Least Sqqguares Filtering With Wiener filtering we have a K as a constant of the ratio](https://reader033.vdocuments.us/reader033/viewer/2022041702/5e4286345a1b89656600351a/html5/thumbnails/68.jpg)
Disadvantage-X-ray scatter, higher doses than G1 d G2 t hi bl SNRand G2 to achieve comparable SNR
Fifth generation (G5)CT scanners-electron beam CT li i t ll h i l ti b l i
68
eliminates all mechanical motion by employing electron beams controlled electromagnetically
Sixth generation CT (helical CT) G3 or G4 scanner Sixth generation CT (helical CT)-G3 or G4 scanner is configured using slip rings that eliminate the need for electrical and signal cabling between the for electrical and signal cabling between the source/detectors and the processing unit
The source/detector pair rotates cntinuously through The source/detector pair rotates cntinuously through 360 deg while the patient is moved at a constant speed along the axis perpendicular to the scanp g p p
Results in a continuous helical volume of data that is processed to obtain individual slice images
INEL 5327 ECE, UPRM
![Page 69: I RImage Restoration Chapter 5 Prof. Vidya Manian Depppggt ...manian/chapter5IP.pdfConstrained Least Sqqguares Filtering With Wiener filtering we have a K as a constant of the ratio](https://reader033.vdocuments.us/reader033/viewer/2022041702/5e4286345a1b89656600351a/html5/thumbnails/69.jpg)
Seventh generataion (G7) CT –multislice CT
69
Seventh generataion (G7) CT multislice CT scanners –thick fan beams are used in conjunction with parallel banks of detectors to collect volumetric pCT data simultaneously
3D cross sectional slabs rather than single ones are ggenerated per X ray burst
Uses Xray tubes more economically, reducing cost and dosage
INEL 5327 ECE, UPRM
![Page 70: I RImage Restoration Chapter 5 Prof. Vidya Manian Depppggt ...manian/chapter5IP.pdfConstrained Least Sqqguares Filtering With Wiener filtering we have a K as a constant of the ratio](https://reader033.vdocuments.us/reader033/viewer/2022041702/5e4286345a1b89656600351a/html5/thumbnails/70.jpg)
70
INEL 5327 ECE, UPRM
![Page 71: I RImage Restoration Chapter 5 Prof. Vidya Manian Depppggt ...manian/chapter5IP.pdfConstrained Least Sqqguares Filtering With Wiener filtering we have a K as a constant of the ratio](https://reader033.vdocuments.us/reader033/viewer/2022041702/5e4286345a1b89656600351a/html5/thumbnails/71.jpg)
The 71
INEL 5327 ECE, UPRM
![Page 72: I RImage Restoration Chapter 5 Prof. Vidya Manian Depppggt ...manian/chapter5IP.pdfConstrained Least Sqqguares Filtering With Wiener filtering we have a K as a constant of the ratio](https://reader033.vdocuments.us/reader033/viewer/2022041702/5e4286345a1b89656600351a/html5/thumbnails/72.jpg)
Geometric Mean Filter
Generalizing the Wiener Filter we obtain the Generalizing the Wiener Filter we obtain the following expression:
with α and β being positive real constantswith α and β being positive, real constants
INEL 5327 ECE, UPRM72
![Page 73: I RImage Restoration Chapter 5 Prof. Vidya Manian Depppggt ...manian/chapter5IP.pdfConstrained Least Sqqguares Filtering With Wiener filtering we have a K as a constant of the ratio](https://reader033.vdocuments.us/reader033/viewer/2022041702/5e4286345a1b89656600351a/html5/thumbnails/73.jpg)
Geometric Transformations
Modify the spatial relationships between pixels in an y p p pimage.
In this type of transformation we have to approach:
S l f f- Spatial transformation, which defines the “rearrangement” of pixels
- Gray – level interpolation which deals with the - Gray level interpolation which deals with the assignment of gray levels to pixels in the spatially transformed image.
INEL 5327 ECE, UPRM73
![Page 74: I RImage Restoration Chapter 5 Prof. Vidya Manian Depppggt ...manian/chapter5IP.pdfConstrained Least Sqqguares Filtering With Wiener filtering we have a K as a constant of the ratio](https://reader033.vdocuments.us/reader033/viewer/2022041702/5e4286345a1b89656600351a/html5/thumbnails/74.jpg)
Exercises
5.11 Refer to the contraharmonic filter in Eq. 5.3-6,
74
5.11 Refer to the contraharmonic filter in Eq. 5.3 6, Explain why the filter is effective in eliminating pepper noise when Q is positive.p pp p
Explain why the filter is effective in eliminating salt noise when Q is negative.g
Discuss the behavior of the filter when Q=-1 Discuss behavior of filter in areas of constant Discuss behavior of filter in areas of constant
intensity levels 5.10) Explain why subimage obtained with 5.10) Explain why subimage obtained with
geometric mean filtering is less blurredINEL 5327 ECE, UPRM
![Page 75: I RImage Restoration Chapter 5 Prof. Vidya Manian Depppggt ...manian/chapter5IP.pdfConstrained Least Sqqguares Filtering With Wiener filtering we have a K as a constant of the ratio](https://reader033.vdocuments.us/reader033/viewer/2022041702/5e4286345a1b89656600351a/html5/thumbnails/75.jpg)
A professor of archeology doing research on
75
A professor of archeology doing research on currency exchange, finds that the photos of the coins are blurred to the point where the date and other pmarkings are not readable. The blurring is due to the camera being out of focus when the pictures were taken. Propose a step-by-steps solution to restore the images to the point that the Professor
d th ki i th d f can read the markings, given the camera used for taking the photos and also the coins.
INEL 5327 ECE, UPRM
![Page 76: I RImage Restoration Chapter 5 Prof. Vidya Manian Depppggt ...manian/chapter5IP.pdfConstrained Least Sqqguares Filtering With Wiener filtering we have a K as a constant of the ratio](https://reader033.vdocuments.us/reader033/viewer/2022041702/5e4286345a1b89656600351a/html5/thumbnails/76.jpg)
Restoring images degraded due to linear motion (in x-or y direction) and circular motionor y- direction) and circular motion
5.17) Linear motion. Estimate blurring function H(u,v)76
5.19) Images from a craft are blurred due to circular motion (rapid rotation of the craft about its vertical axis). The images are blurred by uniform rotational motion. The craft rotation was limited to /8 radians d Th during image acquisition. The image acquisition process can be modeled as an ideal shutter that is open only during the time the craft rotated the /8 open only during the time the craft rotated the /8 radians. Assuming vertical motion was negligible during image acquisition, formulate a solution for during image acquisition, formulate a solution for restoring the images.
INEL 5327 ECE, UPRM