Digital Image Processing

MODULE 2

[Link]
By
Shyno K G
(Asst. Professor, Dept. Of ECE)

For more study materials: [Link]

 Review of Matrix Theory
Basic Image Operations through Matrices
The following 2D-Operations are covered:
➢ 2D Linear Convolution
➢ 2D Circular Convolution
➢ 2D Linear Correlation (Auto/Cross)
➢ 2D Circular Correlation (Auto/Cross)

[Link]
1. 2D Linear Convolution
➢ The convolution between x[m,n] and h[m,n] done by forming Block Toeplitz
matrix out of one matrix and multiplying it with the column matrix formed by
the other
➢ Block matrices are denoted by and the number of matrices are
determined by m and n

1/22/2018 MOD 2 2
For more study materials: [Link]
 Review of Matrix Theory
2D Linear Convolution-Algorithm

Problems:
(Refer Note book)

[Link]

1/22/2018 MOD 2 3
For more study materials: [Link]
 Review of Matrix Theory
2. 2D Circular Convolution
➢ The circular convolution between x[m,n] and h[m,n]
done by forming Block Circulant matrix out of one matrix and multiplying it
with the column matrix formed by the other

[Link]
➢ Application
➢ It can be used for ZOOM operation
➢ Circular convolution interpolates missing values as shown:

1/22/2018 MOD 2 4
For more study materials: [Link]
 Review of Matrix Theory
2D Circular Convolution-Algorithm

Problems:
(Refer Note book)

[Link]

1/22/2018 MOD 2 5
For more study materials: [Link]
 Review of Matrix Theory
3. 2D Linear Correlation
➢ The linear correlation between x[m,n] and h[m,n] is similar to the linear
convolution of one image to the folded version of other image

➢ Form Block Toeplitz matrix from x[m,n] and multiply with column matrix

[Link]
formed from h[-m,-n]
➢ Re arrange the rows we get the O/P

1/22/2018 MOD 2 6
For more study materials: [Link]
 Review of Matrix Theory
2D Linear Correlation-Algorithm

Problems:
(Refer Note book)

[Link]

1/22/2018 MOD 2 7
For more study materials: [Link]
 Review of Matrix Theory
4. 2D Circular Correlation
➢ The Circular correlation between x[m,n] and h[m,n] is similar to the Circular
convolution of one image to the folded version of other image

➢ Form Block Circulant matrix from x[m,n] and multiply with column matrix

[Link]
formed from h[-m,-n]
➢ Re arrange the rows we get the O/P

1/22/2018 MOD 2 8
For more study materials: [Link]
 Review of Matrix Theory
2D Circular Correlation-Algorithm

Problems:
(Refer Note book)

[Link]

1/22/2018 MOD 2 9
For more study materials: [Link]
 2D-Image Transforms
2D-DFT (Discrete Fourier Transform)

[Link]
➢ Inverse 2D DFT is given as:

➢ For an NXN image:

1/22/2018 MOD 2 10
For more study materials: [Link]
 2D-Image Transforms
2D-DFT (Discrete Fourier Transform)

➢ In Polar form:

➢ Magnitude Spectrum:

[Link]
➢ Phase Spectrum:

➢ Frequency Range:
(For Standard Representation)
(For Optical Representation)

1/22/2018 MOD 2 11
For more study materials: [Link]
 2D-Image Transforms
2D-DFT (Discrete Fourier Transform)
➢ 2 Types of representations:
1. Standard Representation
2. Optical Representation

[Link]

1/22/2018 MOD 2 12
For more study materials: [Link]
 2D-Image Transforms
2D-DFT-Properties
1. Separable Property
2. Spatial Shift
3. Periodicity
4. Convolution Property
5. Correlation Property
6.
7. [Link]
Scaling Property
Conjugate Symmetry
8. Orthogonal Property
9. Multiplication by exponents
10. Rotation Property

1/22/2018 MOD 2 13
For more study materials: [Link]
 2D-Image Transforms
2D-DFT-Properties
1. Separable Property

[Link]

1/22/2018 MOD 2 14
For more study materials: [Link]
 2D-Image Transforms
2D-DFT-Properties
1. Separable Property
➢ The main advantage is:
➢ Fourier transform in any dimension can be performed by applying 1D transform
on each direction
2. Spatial Shift

[Link]

1/22/2018 MOD 2 15
For more study materials: [Link]
 2D-Image Transforms
2D-DFT-Properties
2. Spatial Shift

[Link]

1/22/2018 MOD 2 16
For more study materials: [Link]
 2D-Image Transforms
2D-DFT-Properties
2. Spatial Shift
➢ Sub: the above result in eqn (4) we get:

[Link]
3. Periodicity Property

1/22/2018 MOD 2 17
For more study materials: [Link]
 2D-Image Transforms
2D-DFT-Properties
3. Periodicity Property

[Link]

1/22/2018 MOD 2 18
For more study materials: [Link]
 2D-Image Transforms
2D-DFT-Properties
4. Convolution Property

[Link]

1/22/2018 MOD 2 19
For more study materials: [Link]
 2D-Image Transforms
2D-DFT-Properties
4. Convolution Property

[Link]

1/22/2018 MOD 2 20
For more study materials: [Link]
 2D-Image Transforms
2D-DFT-Properties
4. Convolution Property

[Link]

1/22/2018 MOD 2 21
For more study materials: [Link]
 2D-Image Transforms
2D-DFT-Properties
5. Correlation Property(Explained using Separable Property)

[Link]

1/22/2018 MOD 2 22
For more study materials: [Link]
 2D-Image Transforms
2D-DFT-Properties
5. Correlation Property

[Link]

1/22/2018 MOD 2 23
For more study materials: [Link]
 2D-Image Transforms
2D-DFT-Properties
6. Scaling Property

[Link]

1/22/2018 MOD 2 24
For more study materials: [Link]
 2D-Image Transforms
2D-DFT-Properties
6. Scaling Property

7. [Link]
Conjugate Symmetry

1/22/2018 MOD 2 25
For more study materials: [Link]
 2D-Image
2D-DFT-Properties
Transforms
7. Conjugate Symmetry

[Link]
8. Orthogonality Property

1/22/2018 MOD 2 26
For more study materials: [Link]
 2D-Image
2D-DFT-Properties
Transforms
9. Multiplication by Exponential

[Link]

1/22/2018 MOD 2 27
For more study materials: [Link]
 2D-Image
2D-DFT-Properties
Transforms
10. Rotation Property

[Link]

1/22/2018 MOD 2 28
For more study materials: [Link]
 2D-Image Transforms
2D-DFT-Properties (For 5 Marks)

[Link]

➢ Problems refer Notebook:

1/22/2018 MOD 2 29
For more study materials: [Link]
 2D-Image Transforms
Importance of Phase
➢ Consider two images; Convert them into frequency domain using DFT
➢ Interchange their phase
➢ Now find IDFT to reconstruct the images
➢ The images got exchanged as if they where interchanged their magnitudes

[Link]

1/22/2018 MOD 2 30
For more study materials: [Link]
 2D-Image Transforms
Walsh Transform

[Link]

1/22/2018 MOD 2 31
For more study materials: [Link]
 2D-Image Transforms
Walsh Transform

[Link]

1/22/2018 MOD 2 32
For more study materials: [Link]
 2D-Image Transforms
Walsh Transform

[Link]
Walsh Basis for N=4

(Problems Refer Note:)

1/22/2018 MOD 2 33
For more study materials: [Link]
 2D-Image Transforms
Hadamard Transform

[Link]

1/22/2018 MOD 2 34
For more study materials: [Link]
 2D-Image Transforms
Haar Transform
➢ Based on orthogonal matrices whose elements are either [1, -1, 0] multiplied
by powers of
➢ Computationally efficient ( N-point vector requires 2(N-1) additions and N
multiplication

[Link]

1/22/2018 MOD 2 35
For more study materials: [Link]
 2D-Image Transforms
Haar Transform

[Link]

1/22/2018 MOD 2 36
For more study materials: [Link]
 2D-Image Transforms
Haar Transform

➢ Haar matrix for N=2

[Link]

➢ Problems refer Notes

1/22/2018 MOD 2 37
For more study materials: [Link]
 2D-Image Transforms
Discrete Cosine Transform (DCT)
➢ By Ahmed, Natarajan and Rao
➢ The one dimensional DCT Kernel:

[Link]
➢ Where

➢ The one dimensional IDCT Kernel:

1/22/2018 MOD 2 38
For more study materials: [Link]
 2D-Image Transforms
Discrete Cosine Transform (DCT)

[Link]

1/22/2018 MOD 2 39
For more study materials: [Link]
 2D-Image Transforms
DCT-Properties

[Link]

1/22/2018 MOD 2 40
For more study materials: [Link]
 2D-Image Transforms
DCT-Properties

[Link]

1/22/2018 MOD 2 41
For more study materials: [Link]
 2D-Image Transforms
DCT-Base for N=4

[Link]

➢ Problems refer Notes: MOD 2

1/22/2018 For more study materials: [Link] 42
 2D-Image Transforms
Karhunen-Loeve Transform (KL-Transform/ Hotelling
Transform)
➢ Kari Karhunen and Michel Loeve (In Continuous Random Process)
➢ Harold Hotelling (In Discrete Process)
➢ Bases : Orthogonal Eigen vectors of the Covariance matrix of image
➢ Most of the Energy will concentrate on first few components: High Energy

[Link]
Compaction
Drawbacks
➢ Bases depends on I/P. No mathematical structure. So higher computation time
➢ KL transform has multiply/Add operations . DCT/DFT have only

Applications
➢ Clustering Analysis: Depending on the variance of I/P we can reduce the
dimensions of an I/P data called ICA( Independent Component Analysis)
➢ Image Compression: Higher energy compaction results in transform with
largest No. of zeros
➢ 1/22/2018
Problems refer Notes:* MOD 2
For more study materials: [Link] 43
 2D-Image Transforms
Singular Value Decomposition(SVD)
➢ SVD for square matrix is was discovered by Beltrami and Jordan
➢ Extended to rectangular matrices by Eckart
➢ The SVD of a rectangular matrix A of size mxn is given as :

[Link]

1/22/2018 MOD 2 44
For more study materials: [Link]
 2D-Image Transforms
Singular Value Decomposition(SVD)
Relation between SVD and Eigen Values

[Link]
Where, U is the Eigen vector of
And V is the Eigen vector of
Diagonal of D is the square roots of Eigen values
➢ If A is real then singular values, U and V are real
Properties of SVD

1/22/2018 MOD 2 45
For more study materials: [Link]
 2D-Image Transforms
Singular Value Decomposition(SVD)
Applications of SVD

[Link]
Problems Refer Notes*

1/22/2018 MOD 2 46
For more study materials: [Link]