Information Theory Fundamentals: Distance Between Two Images Based On Pixels

You might also like

Download as pdf or txt
Download as pdf or txt
You are on page 1of 24

INFORMATION THEORY

FUNDAMENTALS

Unit IV
Distance between two images based
on pixels
Session Meta Data

Author Dr T Sree Sharmila

Reviewer

Version Number 1.0

Release Date 22.03.2021

2 v 1.0
Agenda
• Introduction
• Neighbor of a pixel
• Absolute difference of two images
• Distance between two images based on pixels
1. The Euclidean distance
2. Cosine distance
3. Frobenius distance
4. Mean square error
5. Image histogram
• Summary
• Test your understanding
• References
3 v 1.0
Introduction
• Images are represented in rows and columns we have
the following syntax in which images are represented:

• Every element of this matrix is called image element,


picture element, or pixel.

4 v 1.0
Neighbor of a pixel
• Let the position of pixel P be the
coordinates (x,y).
• It has two horizontal neighbors and two
vertical neighbors:
• Horizontal neighbors C: (x,y-1) D: (x,y+1)
• Vertical Neighbors: A: (x-1,y) B: (x+1,y)
• These horizontal and vertical neighbors
are called the 4-nighbors of P and the set
is denoted by
• N4(P)={A, B, C, D}={(x-1,y), (x+1,y), (x,y-
1), (x,y+1)}
• If P is on the border of the image, some of
the neighbors may not exist.

5 v 1.0
Neighbor of a pixel
• P also has four diagonal neighbors:
L: (x-1,y-1)
M: (x-1,y+1)
N : (x+1,y-1)
O: (x+1,y+1)
• This set is denoted by ND(P) ={L, M,
N, O}.
• All are together called the 8-
neighbors of P, and are denoted by
N8(P)

6 v 1.0
Absolute difference of two images
• Image difference
• It subtracts each element in array Y from the
corresponding element in array X and returns the
absolute difference in the corresponding element of the
output array Z.

7 v 1.0
Distance between two images based on
pixels
1. The Euclidean distance
2. Cosine distance
3. Frobenius distance
4. Mean square error
5. Image histogram

8 v 1.0
The Euclidean distance
• The distance between two
points in either the plane or
3-dimensional space
measures the length of a
segment connecting the
two points.
• It is the most obvious way
of representing distance
between two points.

9 v 1.0
The Euclidean distance
• If the points (x1,y1) and (x2,y2)
are in 2-dimensional space,
then the Euclidean distance

10 v 1.0
Distance between Test image & training image

2116 144 196 1

6724 169 1521 1089


Square differences,
Sum up, & Take root
144 100 0 900 √26280 = 162.1
4 1024 484 11664

11 v 1.0
Cosine distance for two points (vectors)

• Derived from cosine similarity


• Cosine similarity = Inner product = dot product

p = ( p1 , p2 , p3 ) & q = (q1 , q2 , q3 )
similarity = ( p1.q1 ) + ( p2 .q2 ) + ( p3 .q3 )
 q1 
similarity = p.q T = [ p1 p2 p3 ].q2 
 q3 
• Distance = 1-similarity

12 v 1.0
Frobenius inner product

• Take two matrices and returns a number


• Consider two matrices, A and B of size (mxn)
• Extend the concept of dot product of vectors

• Distance = 1 - similarity

13 v 1.0
e.g.

14 v 1.0
Example application

• Two images are given ≡ Two matrices are given


• Cosine distance measures the distance between
two images – works better than Euclidean

15 v 1.0
560 640 240 306

720 230 11392 13300 Sum up: 177839

288 416 31684 34000

8 0 59415 24640

16 v 1.0
Mean squared error
• Suppose I and K are the original and the noisy images.
• MSE is referred to the error signal, which is the
difference between the original and distorted signals.
• If one of the signals is an original signal (image), and the
other is a distorted version of it whose quality is being
evaluated, then the MSE may also be regarded as a
measure of signal quality.

17 v 1.0
Histogram matching
• It transforms the 2-D grayscale or truecolor
image I returning output image J whose histogram
approximately matches the histogram of the reference
image ref.
• If both I and ref are truecolor RGB images, then Image
histogram matching matches each color channel
of I independently to the corresponding color channel of ref.
• If I is a truecolor RGB image and ref is a grayscale image,
then Image histogram matching matches each channel
of I against the single histogram derived from ref.
• If I is a grayscale image, then ref must also be a grayscale
image.

18 v 1.0
Applications

19 v 1.0
Measuring size of an object

20 v 1.0
Measuring size of an object

21 v 1.0
Summary
• The following distance measures have been studied:
1. The Euclidean distance
2. Cosine distance
3. Frobenius distance
4. Mean square error
5. Image histogram

22 v 1.0
Test your understanding
• If there is a very high peak right at the top end of the
histogram, what does this suggest?
• Suppose that you had a scene of three objects of
different distinct intensities against an extremely bright
background. What would the corresponding histogram
look like?
• What are the different ways to find the distance between
two images based on probability? What are the different
ways to find the distance between two images based on
probability?
• Explain the distance between them by any two distance
measure based on pixels with example.
23 v 1.0
References
• Thomas Cover, Joy Thomas, Elements of Information
Theory , Wiley Inderscience, 2nd Edition, 2006.
• R C Gonzalez, and R E Woods, Digital Image
Processing, Pearson, 2018.

24 v 1.0

You might also like