Prof. Dr. Ayad A.
Al -Ani, Professor of Digital Image Processing
Department of Information & Communication Engineering (DICE), College of Information Engineering (COIE), Al -Nahrain University
Chapter 6- 7
The Mathematical Model of Image Formation
Prof. Dr. Ayad A. Al -Ani, Professor of Digital Image Processing
Department of Information & Communication Engineering (DICE), College of Information Engineering (COIE), Al -Nahrain University
According to sifting theorem, we have:
ܠǡ ܡൌ ܠǡ ܠ ઼ ٔ ܡǡ ܡ (1)
Where : ܠǡ ܡtwo dimensional function, ٔ convolution process
ܑ ܠൌ ܠ0 ܡ ܌ܖ܉ൌ ܡ0
઼ ܠǡ ܡ: Dirac - Delta function, define by: ઼ ܠǡ ܡൌ
܍ܛܑܟ ܚ܍ܐܜܗ
Taking the Fourier transform (FT) of eq. 1, we get :
۴ܠ ܂ǡ ܡ ൌ ۴ ܂ሾ ܠǡ ܠ ઼ ٔ ܡǡ ܡሿ
۴ܠ ܂ǡ ܡ ൌ ۴ ܂ሾ ܠǡ ܡሿ . ۴ ܂ሾ઼ ܠǡ ܡሿ
۴ ܝǡ ܞൌ ۴ ܝǡ ܞǤ
Where : ۴ ܝǡ ܞൌ ۴ܠ ܂ǡ ܡ
۴ܠ ઼ ܂ǡ ܡ ൌ i. e.
۴ܠ ઼ ܂ǡ ܡ ൌ ۱܍܉ܕܑ ܍ܐܜ ܖܑ ܔ܍ܠܑܘ ܚܗ ܜܖܑܗܘ ܡܖ܉ ܜ܉ ܜܖ܉ܜܛܖܗ
Prof. Dr. Ayad A. Al -Ani, Professor of Digital Image Processing
Department of Information & Communication Engineering (DICE), College of Information Engineering (COIE), Al -Nahrain University
Now, eq. is given by:
ܠǡ ܡൌ ܠǡ ܠ ઼ ٔ ܡǡ ܡ (1)
Then the mathematical representation of eq. (1) is given by:
ାஶ ାஶ
(ܠǡ ܡሻ ൌ ିஶ ିஶ હǡ ઼ ܠെ હ ǡ ܡെ ܌હ ܌ (2)
We know that:
ሺܠǡ ܡሻ ൌ ℋ ሾሺܠǡ ܡሻሿ (3)
Where: (ܠǡ ܡሻ output (image) function, (ܠǡ ܡሻ input (object) function
ℋ : represent any Operator or impulse function or transform between input
and output .
Prof. Dr. Ayad A. Al -Ani, Professor of Digital Image Processing
Department of Information & Communication Engineering (DICE), College of Information Engineering (COIE), Al -Nahrain University
The response ℋ may be:
Camera
Lens
Optical System
Microscope
Oscilloscope
Voltmeter, Ammeter, «
Telescope
ℋ , also, may be:
Linear system
Non linear system
Prof. Dr. Ayad A. Al -Ani, Professor of Digital Image Processing
Department of Information & Communication Engineering (DICE), College of Information Engineering (COIE), Al -Nahrain University
Linear system must have two properties :
First : Distributivity
ℋ ܠǡ ܡ ܠǡ ܡ ൌ ℋ ܠǡ ܡ ℋ ܠǡ ܡ (4)
Second: homogeneity
ℋ હ ܠǡ ܡ ൌ હℋ ܠǡ ܡ (5)
Where: હ is constant value.
Then in general:
ℋ હ ܠǡ ܡ ܠǡ ܡ ൌ ℋ હ ܠǡ ܡ ℋ ܠǡ ܡ
ℋ હ ܠǡ ܡ ܠǡ ܡ ൌ હ ℋ ܠǡ ܡ ℋ ܠǡ ܡ
ܠǡ ܡൌ હ ܌ ܠǡ ܡ ܌ ܠǡ ܡ
ܠǡ ܡൌ ܠǡ ܡ ܠǡ ܡ
ܠǡ ܡൌ ܠǡ ܡ
Prof. Dr. Ayad A. Al -Ani, Professor of Digital Image Processing
Department of Information & Communication Engineering (DICE), College of Information Engineering (COIE), Al -Nahrain University
Return to eqs. 1 and 3,
ܠǡ ܡൌ ܠǡ ܠ ઼ ٔ ܡǡ ܡ
ሺܠǡ ܡሻ ൌ ℋ ሾ ሺܠǡ ܡሻ ሿ
ܠǡ ܡൌ ℋ ܠǡ ܠ ઼ ٔ ܡǡ ܡ
ାஶ ାஶ
ܠǡ ܡൌ ℋ ିஶ ିஶ હǡ ઼ ܠെ હ ǡ ܡെ ܌હ ܌
Assuming: H is Linear function, i.e.
ℋ હ ܠǡ ܡ ൌ હ ℋ ܠǡ ܡ
since ܠǡ ܡrepresent the intensity equal to constant, then we get ;
ାஶ ାஶ
ܠǡ ܡൌ ିஶ ିஶ હǡ Ǥ ℋ ઼ ܠെ હ ǡ ܡെ ܌હ ܌
Let: ℋ ઼ ܠെ હ ǡ ܡെ ൌ ܐሺܠǢ હǡ ܡǢ ሻ
Then the above equation becomes:
ାஶ ାஶ
ܠǡ ܡൌ ିஶ ିஶ હǡ ܠ ܐǢ હǡ ܡǢ ܌હ ܌ (6)
Prof. Dr. Ayad A. Al -Ani, Professor of Digital Image Processing
Department of Information & Communication Engineering (DICE), College of Information Engineering (COIE), Al -Nahrain University
Where ܠ ܐǢ હ ǡ ܡǢ represent the new impulse function .
This means that ܠ ܐǢ હ ǡ ܡǢ depends on 4 parameters, which is very
complex to treating mathematically .
There are two type of impulse function ;
Impulse function depend on the position of Pixel of the object
Impulse function does not depend on the position of Pixel of the
object
The impulse function can be named:
Blurring function (BF ) or Point Spread Function (PSF)
i.e. ܠ ܐǢ હ ǡ ܡǢ ൌ ܁۾۴ ܠǢ હ ǡ ܡǢ
PSF: determine the energy distribution in the image plane for the point
source Located in the object plane
Prof. Dr. Ayad A. Al -Ani, Professor of Digital Image Processing
Department of Information & Communication Engineering (DICE), College of Information Engineering (COIE), Al -Nahrain University
Then according to the type of PSF, whether it is dependent or independent
on the position of the object ·
s pixel, i. e:
1. Point Spread Function which is Invariant with the Position (Space),
labeled by: SIPSF
2. Point Spread Function which is Variant with the Position (Space),
labeled by: SVPSF
i.e .
SIPSF : Space Invariant Point Spread Function, defined as:
is the PSF that don·
t change its shape with the position of image·
s
pixel
SVPSF: Space Variant Point Spread Function, define as:
is the PSF that change its shape with the position of image·
s pixel .
Prof. Dr. Ayad A. Al -Ani, Professor of Digital Image Processing
Department of Information & Communication Engineering (DICE), College of Information Engineering (COIE), Al -Nahrain University
Now assuming that PSF is space invariant function ; i. e.
ܠ ܐǢ હ ǡ ܡǢ does not depends on the position of the pixel, i. e.
ܠ ܐǢ હ ǡ ܡǢ = ܠ ܐെ હ ǡ ܡെ
This mean that the Blurring Function depend only on 2 parameters [ ሺܠ
Prof. Dr. Ayad A. Al -Ani, Professor of Digital Image Processing
Department of Information & Communication Engineering (DICE), College of Information Engineering (COIE), Al -Nahrain University
Now; assuming that there is an additive uncorrelated noise ܖሺܠǡ ܡሻ in the
optical system, then eq. (8) becomes:
ܠǡ ܡൌ ܠǡ ܠ ܐ ٔ ܡǡ ܡ ܖሺܠǡ ܡሻ (10)
Equation (10 ) is known as:
Mathematical Model of Image Formation
or for simplicity it is named:
Image Formation Model
Now, for Image Formation Model, there are three cases, these are :
Prof. Dr. Ayad A. Al -Ani, Professor of Digital Image Processing
Department of Information & Communication Engineering (DICE), College of Information Engineering (COIE), Al -Nahrain University
First : in case of there is no noise, eq. (10 ) becomes:
ܠǡ ܡൌ ܠǡ ܠ ܐ ٔ ܡǡ ܡ
Here: ܠǡ ܡis : called Blurred image
Second: in case of there is no blurring function, eq. (10 ) becomes:
ܠǡ ܡൌ ܠǡ ܡ ܖሺܠǡ ܡሻ
Here: ܠǡ ܡis : called Noisy image
Third : in case of there is two types of degradation parameters, i. e.
blurring function and noise function, eq. (10 ) becomes:
ܠǡ ܡൌ ܠǡ ܠ ܐ ٔ ܡǡ ܡ ܖሺܠǡ ܡሻ
Here : ܠǡ ܡis : called Degraded image
Prof. Dr. Ayad A. Al -Ani, Professor of Digital Image Processing
Department of Information & Communication Engineering (DICE), College of Information Engineering (COIE), Al -Nahrain University
Return to Eq. (10 ):
ܠǡ ܡൌ ܠǡ ܠ ܐ ٔ ܡǡ ܡ ܖሺܠǡ ܡሻ (10)
Now, taking the Fourier Transform of eq. (10 ), we get :
۴ܠ ܂ǡ ܡൌ ܠǡ ܠ ܐ ٔ ܡǡ ܡ ܖሺܠǡ ܡሻ
۴ܠ ܂ǡ ܡ = ۴ܠ ܂ǡ ܠ ܐ ٔ ܡǡ ܡ ܖሺܠǡ ܡሻ
۴ܠ ܂ǡ = ܡ۴ܠ ܂ǡ ܠ ܐ ٔ ܡǡ ܡ + ۴ܖ ܂ሺܠǡ ܡሻ
۴ܠ ܂ǡ = ܡ۴ܠ ܂ǡ ܡ . ۴ܠ ܐ ܂ǡ ܡ + ۴ܖ ܂ሺܠǡ ܡሻ
۵ ܝǡ ܞൌ ۴ ܝǡ ܞǤ ۶ ܝǡ ܞ ۼሺܝǡ ܞሻ (11)
Comparing between eqs. 10 and 11 , we see that the convolution process in the
spatial domain becomes simple multiplication process in the frequency domain.
Prof. Dr. Ayad A. Al -Ani, Professor of Digital Image Processing
Department of Information & Communication Engineering (DICE), College of Information Engineering (COIE), Al -Nahrain University
Where:
۵ ܝǡ ܞൌ ۴ܠ ܂ǡ ܡ ൌ ۴܂ሾ ܑܠ ܕǡ ܡሿ
۴ ܝǡ ܞൌ ۴ܠ ܂ǡ ܡ ൌ ۴܂ሾ ܠ ܒ܊ܗǡ ܡሿ
۶ ܝǡ ܞൌ ۴ܠ ܐ ܂ǡ ܡ ൌ ۴܂ሾ ܁۾۴ ܠǡ ܡሿ
ܝ ۼǡ ܞൌ ۴ܠ ܖ ܂ǡ ܡ
۵ ܝǡ ܞ: is the Fourier transform of the image
۴ ܝǡ ܞ: is the Fourier transform of the Object
۶ ܝǡ ܞ: is the Fourier transform of the Blurring Function
ܝ ۼǡ ܞ: is the Fourier transform of the Noise
Where : ܠand ܡis known as the Spatial Coordinates
x = 1, 2, 3, …. M y = 1, 2, 3, …. N
While : ܝand ܞis known as the Spatial Frequency Coordinates
u = 1, 2, 3, …. M v = 1, 2, 3, …. N
Prof. Dr. Ayad A. Al -Ani, Professor of Digital Image Processing
Department of Information & Communication Engineering (DICE), College of Information Engineering (COIE), Al -Nahrain University
۶ ܝǡ ܞ: is the Fourier transform of the Blurring Function (PSF), known as
the : Transfer Function ´TF ´
the properties of the Normalized TF are:
1- ܂۴ ǡ ൌ
2- ܂۴ ܝǡ ܞ ܂۴ ǡ
3- ܂۴ ܝǡ ܞൌ ܂۴ െܝǡ െܞ
Prof. Dr. Ayad A. Al -Ani, Professor of Digital Image Processing
Department of Information & Communication Engineering (DICE), College of Information Engineering (COIE), Al -Nahrain University
Degradation function
First ²Blurring Function ( ܠ ܐǡ ) ܡ
Blurring function may be due to :
1. Defects of the Lenses
2. Relative Motion between Object and Optical System
3. Medium between Object and Optical System
4. Exposure Time
Blurring Function means Averaging Function
H. W. Present each of the above Blurring Function Mathematically
Prof. Dr. Ayad A. Al -Ani, Professor of Digital Image Processing
Department of Information & Communication Engineering (DICE), College of Information Engineering (COIE), Al -Nahrain University
1- Aberration of the Lenses may be :
Spherical aberration
Defocusing
Astigmatism
Coma
Field of Curvature
In general, and for simplicity, blurring function is given by Gaussian
function ۵ ܠǡ ܡ:
ܠିܠ ା ܡିܡ Τો
۵ ܠǡ ܡൌ ି܍
ૈఙ మ
All these 3 equations
Τો
۵ ܠǡ ܡൌ ࢠି܍
are Same
ૈఙ మ
ૄିܢ Τો
۵ ܠǡ ܡൌ ି܍
ૈమ
ૄିܚ Τો
۵ ܚൌ ି܍
ૈమ
Prof. Dr. Ayad A. Al -Ani, Professor of Digital Image Processing
Department of Information & Communication Engineering (DICE), College of Information Engineering (COIE), Al -Nahrain University
Where :
μ : is the mean of the Gaussian Function
ķ: is the mean and standard deviation of the Gaussian Function,
r : is the distance from the center of the function ܠ ǡ ܡ
Prof. Dr. Ayad A. Al -Ani, Professor of Digital Image Processing
Department of Information & Communication Engineering (DICE), College of Information Engineering (COIE), Al -Nahrain University
Prof. Dr. Ayad A. Al -Ani, Professor of Digital Image Processing
Department of Information & Communication Engineering (DICE), College of Information Engineering (COIE), Al -Nahrain University
2- Relative motion between object and Optical System may be:
Horizontal
Vertical
Diagonal
Linear
Non- Linear
The FT of the blurring function due to 2D linear motion is given by:
܂
۶ ܝǡ ܞൌ Ǥ ܖܑܛሺૈ܀ሻ ܀ૈܑି܍
ૈ܀
܀ൌ ܉ܝ ܊ܞ
H.W . : Find the blurring function in spatial domain, using 1Dimension?
Prof. Dr. Ayad A. Al -Ani, Professor of Digital Image Processing
Department of Information & Communication Engineering (DICE), College of Information Engineering (COIE), Al -Nahrain University
3- Optical medium between object and optical system may be:
Homogeneous medium
Inhomogeneous medium
4- Exposure time may be:
Short Exposure Speckly PSF
Long Exposure Gaussian Distribution PSF
H. W.: Present the image formation model in case of
Long- Exposure.
Speckly PSF
Prof. Dr. Ayad A. Al -Ani, Professor of Digital Image Processing
Department of Information & Communication Engineering (DICE), College of Information Engineering (COIE), Al -Nahrain University
Second - Noise Function ܠ ܖǡ ܡ
There are different type of the noise function, these types are :
Noise Signal dependent
Noise Signal independent
Also there are :
Additive Noise
Multiplicative Noise
Also there are : different type of the noise Distribution, these types are :
Prof. Dr. Ayad A. Al -Ani, Professor of Digital Image Processing
Department of Information & Communication Engineering (DICE), College of Information Engineering (COIE), Al -Nahrain University
type of the noise Distribution
1. Gaussian Noise
2. Uniform Noise
3. Impulse Noise (Salt and Pepper Noise)
4. Erlang Noise (Gamma Noise )
5. Poisson Noise
6. Periodic Noise
7. Rayleigh Noise
8. Exponential Noise
Prof. Dr. Ayad A. Al -Ani, Professor of Digital Image Processing
Department of Information & Communication Engineering (DICE), College of Information Engineering (COIE), Al -Nahrain University
Probability Density Function (PDF)
The PDF is used to specify the probability of the Random
Variable falling within a particular range of values, as instead of taking on
any one value.
. بدال من أخذ أي قيمة واحدة، لتحديد احتمال وقوع المتغير العشوائي ضمن نطاق معين من القيمPDF يتم استخدام
This probability is given by the integral of this variable's PDF over that
range, i. e. it is given by the area under the density function but above the
horizontal axis and between the lowest and greatest values of the range.
The probability density function is nonnegative everywhere, and its
Normalized integral over the entire space is equal to 1.
Prof. Dr. Ayad A. Al -Ani, Professor of Digital Image Processing
Department of Information & Communication Engineering (DICE), College of Information Engineering (COIE), Al -Nahrain University
1- Gaussian Noise
Gaussian noise is used as additive White Gaussian noise. In other words,
the values that the noise can take are Gaussian- distributed .
Used frequently in practice
Principal sources of Gaussian noise in digital images arise during
acquisition e. g. sensor noise caused by poor illumination, high
temperature, and transmission, e. g. electronic circuit noise. Also, in
Telecommunications and Computer Networking, communication channels
can be affected by Gaussian noise coming from many natural sources.
This type of noise adds more Noise to the mid tones and less noise to
the shadows and highlight regions of the image
named after Carl Friedrich Gauss.
Prof. Dr. Ayad A. Al -Ani, Professor of Digital Image Processing
Department of Information & Communication Engineering (DICE), College of Information Engineering (COIE), Al -Nahrain University
Gaussian noise
The PDF ( ) ܢ ۾of the Gaussian Noise Distribution is given by:
ૄିܢ Τો
ܢ ۾ൌ ି܍
ૈమ
ାஶ
ିஶ ܢ܌ ܢ ۾ൌ
Where: Ăis the mean of the noise function,
ķ2 is the variance of the noise function
Prof. Dr. Ayad A. Al -Ani, Professor of Digital Image Processing
Department of Information & Communication Engineering (DICE), College of Information Engineering (COIE), Al -Nahrain University
2- Uniform noise
The noise caused by quantizing the pixels of a sensed image to a No. of
discrete levels is known as quantization noise.
It has an approximately uniform distribution, i. e. adds random colour
noise of equal intensity all over the image.
The effect is noticeably more subtle than Gaussian Noise .
Less practical, used for random number generator
Prof. Dr. Ayad A. Al -Ani, Professor of Digital Image Processing
Department of Information & Communication Engineering (DICE), College of Information Engineering (COIE), Al -Nahrain University
Uniform noise
The PDF ( ) ܢ ۾of the Uniform Noise Distribution is given by:
܉ ܚܗ ܈ ܊
܈ ۾ൌ ൦ ܊െ ܉
܍ܛܑܟܚ܍ܐܜܗ
܉ା܊ ܉ି܊
ܖ܉܍ܕൌ ૄ ൌ ܍܋ܖ܉ܑܚ܉ܞൌ ો ൌ
Histogram of uniform
Prof. Dr. Ayad A. Al -Ani, Professor of Digital Image Processing
Department of Information & Communication Engineering (DICE), College of Information Engineering (COIE), Al -Nahrain University
