21CSE251T - Digital Image Processing

Dr. Hariprasad.S
Asst. Professor
Dept. of CINTEL

30-01-2025 Dr. HARIPRASAD S, AP/CINTEL

 Objective

To provide background material in support of topics in Digital Image Processing that

are based on matrices and/or vectors.
 Some Definitions

An m×n (read "m by n") matrix, denoted by A, is a rectangular array of entries or

elements (numbers, or symbols representing numbers) enclosed typically by
square brackets, where m is the number of rows and n the number of columns.
 Definitions (Con’t)

• A is square if m= n.
• A is diagonal if all off-diagonal elements are 0, and not all
diagonal elements are 0.
• A is the identity matrix ( I ) if it is diagonal and all diagonal
elements are 1.
• A is the zero or null matrix ( 0 ) if all its elements are 0.
• The trace of A equals the sum of the elements along its main
diagonal.
• Two matrices A and B are equal iff the have the same
number of rows and columns, and aij = bij .
 Definitions (Con’t)

• The transpose AT of an m×n matrix A is an n×m matrix

obtained by interchanging the rows and columns of A.
• A square matrix for which AT=A is said to be symmetric.
• Any matrix X for which XA=I and AX=I is called the
inverse of A.
• Let c be a real or complex number (called a scalar). The
scalar multiple of c and matrix A, denoted cA, is obtained
by multiplying every elements of A by c. If c = −1, the
scalar multiple is called the negative of A.
 Definitions (Con’t)

A column vector is an m × 1 matrix:

A row vector is a 1 × n matrix:

A column vector can be expressed as a row vector by using

the transpose:
 Some Basic Matrix Operations

• The sum of two matrices A and B (of equal dimension),

denoted A + B, is the matrix with elements aij + bij.
• The difference of two matrices, A− B, has elements aij − bij.
• The product, AB, of m×n matrix A and p×q matrix B, is an
m×q matrix C whose (i,j)-th element is formed by multiplying
the entries across the ith row of A times the entries down the
jth column of B; that is,
 Some Basic Matrix Operations (Con’t)

The inner product (also called dot product) of two vectors

is defined as

Note that the inner product is a scalar.

 Vectors and Vector Spaces

A vector space is defined as a nonempty set V of entities called vectors

and associated scalars that satisfy the conditions outlined in A through
C below. A vector space is real if the scalars are real numbers; it is
complex if the scalars are complex numbers.
• Condition A: There is in V an operation called vector addition,
denoted x + y, that satisfies:
1. x + y = y + x for all vectors x and y in the space.
2. x + (y + z) = (x + y) + z for all x, y, and z.
3. There exists in V a unique vector, called the zero vector, and
denoted 0, such that x + 0 = x and 0 + x = x for all vectors x.
4. For each vector x in V, there is a unique vector in V, called
the negation of x, and denoted −x, such that x + (− x) = 0 and
(− x) + x = 0.
 Vectors and Vector Spaces (Con’t)

• Condition B: There is in V an operation called multiplication by a

scalar that associates with each scalar c and each vector x in V a
unique vector called the product of c and x, denoted by cx and xc,
and which satisfies:
1. c(dx) = (cd)x for all scalars c and d, and all vectors x.
2. (c + d)x = cx + dx for all scalars c and d, and all vectors x.
3. c(x + y) = cx + cy for all scalars c and all vectors x and y.

• Condition C: 1x = x for all vectors x.

 Vectors and Vector Spaces (Con’t)

We are interested particularly in real vector spaces of real m×1 column matrices.
We denote such spaces by m , with vector addition and multiplication by scalars
being as defined earlier for matrices. Vectors (column matrices) in m are written as
 Vectors and Vector Spaces (Con’t)

Example
The vector space with which we are most familiar is the two-dimensional real vector
space 2 , in which we make frequent use of graphical representations for operations
such as vector addition, subtraction, and multiplication by a scalar. For instance,
consider the two vectors

Using the rules of matrix addition and subtraction we have

 Vectors and Vector Spaces (Con’t)

Example (Con’t)
The following figure shows the familiar graphical representation of the preceding
vector operations, as well as multiplication of vector a by scalar c = −0.5.
 Vectors and Vector Spaces (Con’t)

Consider two real vector spaces V0 and V such that:

• Each element of V0 is also an element of V (i.e., V0 is a subset
of V).
• Operations on elements of V0 are the same as on elements of
V. Under these conditions, V0 is said to be a subspace of V.
A linear combination of v1,v2,…,vn is an expression of the form

where the ’s are scalars.

 Vectors and Vector Spaces (Con’t)

A vector v is said to be linearly dependent on a set, S, of vectors v1,v2,…,vn if and only if

v can be written as a linear combination of these vectors. Otherwise, v is linearly
independent of the set of vectors v1,v2,…,vn .
 Vectors and Vector Spaces (Con’t)

A set S of vectors v1,v2,…,vn in V is said to span some subspace V0 of V if and only if S is a

subset of V0 and every vector v0 in V0 is linearly dependent on the vectors in S. The set
S is said to be a spanning set for V0. A basis for a vector space V is a linearly
independent spanning set for V. The number of vectors in the basis for a vector space is
called the dimension of the vector space. If, for example, the number of vectors in the
basis is n, we say that the vector space is n-dimensional.
 Vectors and Vector Spaces (Con’t)

An important aspect of the concepts just discussed lies in the representation of

any vector in m as a linear combination of the basis vectors. For example, any
vector

in 3 can be represented as a linear combination of the basis

vectors
 Vector Norms

A vector norm on a vector space V is a function that assigns to each vector v in V a

nonnegative real number, called the norm of v, denoted by ||v||. By definition,
the norm satisfies the following conditions:
 Vector Norms (Con’t)

There are numerous norms that are used in practice. In our work, the norm most
often used is the so-called 2-norm, which, for a vector x in real m, space is defined
as

which is recognized as the Euclidean distance from the origin to point x; this gives the
expression the familiar name Euclidean norm. The expression also is recognized as
the length of a vector x, with origin at point 0. From earlier discussions, the norm
also can be written as
 Vector Norms (Con’t)

The Cauchy-Schwartz inequality states that

Another well-known result used in the book is the expression

where  is the angle between vectors x and y. From these expressions it follows
that the inner product of two vectors can be written as

Thus, the inner product can be expressed as a function of the norms of the
vectors and the angle between the vectors.
 Vector Norms (Con’t)

From the preceding results, two vectors in m are orthogonal if and only if their
inner product is zero. Two vectors are orthonormal if, in addition to being
orthogonal, the length of each vector is 1.

From the concepts just discussed, we see that an arbitrary vector a is turned into a
vector an of unit length by performing the operation an = a/||a||. Clearly, then,
||an|| = 1.

A set of vectors is said to be an orthogonal set if every two vectors in the set are
orthogonal. A set of vectors is orthonormal if every two vectors in the set are
orthonormal.
 Some Important Aspects of Orthogonality

Let B = {v1,v2,…,vn } be an orthogonal or orthonormal basis in the sense defined in

the previous section. Then, an important result in vector analysis is that any vector
v can be represented with respect to the orthogonal basis B as

where the coefficients are given by

 Orthogonality (Con’t)

The key importance of this result is that, if we represent a vector as a linear

combination of orthogonal or orthonormal basis vectors, we can determine the
coefficients directly from simple inner product computations. It is possible to
convert a linearly independent spanning set of vectors into an orthogonal
spanning set by using the well-known Gram-Schmidt process. There are
numerous programs available that implement the Gram-Schmidt and similar
processes, so we will not dwell on the details here.
 Eigenvalues & Eigenvectors

Definition: The eigenvalues of a real matrix M are the real numbers  for which
there is a nonzero vector e such that
Me =  e.
The eigenvectors of M are the nonzero vectors e for which there is a real number
 such that Me =  e.

If Me =  e for e  0, then e is an eigenvector of M associated with eigenvalue ,

and vice versa. The eigenvectors and corresponding eigenvalues of M constitute
the eigensystem of M.

Numerous theoretical and truly practical results in the application of matrices and
vectors stem from this beautifully simple definition.
 Eigenvalues & Eigenvectors (Con’t)

Example: Consider the matrix

and

In other words, e1 is an eigenvector of M with associated

eigenvalue 1, and similarly for e2 and 2.
 Eigenvalues & Eigenvectors (Con’t)

The following properties, which we give without proof, are essential background in
the use of vectors and matrices in digital image processing. In each case, we
assume a real matrix of order m×m although, as stated earlier, these results are
equally applicable to complex numbers.

1. If {1, 2,…, q, q  m, is set of distinct eigenvalues of M, and

ei is an eigenvector of M with corresponding eigenvalue i, i
= 1,2,…,q, then {e1,e2,…,eq} is a linearly independent set of
vectors. An important implication of this property: If an m×m
matrix M has m distinct eigenvalues, its eigenvectors will
constitute an orthogonal (orthonormal) set, which means that
any m-dimensional vector can be expressed as a linear
combination of the eigenvectors of M.
 Eigenvalues & Eigenvectors (Con’t)

2. The numbers along the main diagonal of a diagonal matrix

are equal to its eigenvalues. It is not difficult to show
using the definition Me =  e that the eigenvectors can be
written by inspection when M is diagonal.

3. A real, symmetric m×m matrix M has a set of m linearly

independent eigenvectors that may be chosen to form an
orthonormal set. This property is of particular importance
when dealing with covariance matrices (e.g., see Section
11.4 and our review of probability) which are real and
symmetric.
 Eigenvalues & Eigenvectors (Con’t)

4. A corollary of Property 3 is that the eigenvalues of an m×m real

symmetric matrix are real, and the associated eigenvectors may
be chosen to form an orthonormal set of m vectors.

5. Suppose that M is a real, symmetric m×m matrix, and that we

form a matrix A whose rows are the m orthonormal eigenvectors
of M. Then, the product AAT=I because the rows of A are
orthonormal vectors. Thus, we see that A−1= AT when matrix A
is formed in the manner just described.

6. Consider matrices M and A in 5. The product D = AMA−1 =

AMAT is a diagonal matrix whose elements along the main
diagonal are the eigenvalues of M. The eigenvectors of D are
the same as the eigenvectors of M.
 Eigenvalues & Eigenvectors (Con’t)

Example
Suppose that we have a random population of vectors, denoted by {x}, with
covariance matrix (see the review of probability):

Suppose that we perform a transformation of the form y = Ax on each vector x, where

the rows of A are the orthonormal eigenvectors of Cx. The covariance matrix of the
population {y} is
 Eigenvalues & Eigenvectors (Con’t)

From Property 6, we know that Cy=ACxAT is a diagonal matrix with the eigenvalues of Cx
along its main diagonal. The elements along the main diagonal of a covariance matrix
are the variances of the components of the vectors in the population. The off diagonal
elements are the covariances of the components of these vectors.
The fact that Cy is diagonal means that the elements of the vectors in the population {y}
are uncorrelated (their covariances are 0). Thus, we see that application of the linear
transformation y = Ax involving the eigenvectors of Cx decorrelates the data, and the
elements of Cy along its main diagonal give the variances of the components of the y's
along the eigenvectors. Basically, what has
 Eigenvalues & Eigenvectors (Con’t)

been accomplished here is a coordinate transformation that aligns the data along
the eigenvectors of the covariance matrix of the population.
The preceding concepts are illustrated in the following figure. Part (a) shows a
data population {x} in two dimensions, along with the eigenvectors of Cx (the black
dot is the mean). The result of performing the transformation y=A(x − mx) on the
x's is shown in Part (b) of the figure.
The fact that we subtracted the mean from the x's caused the y's to have zero
mean, so the population is centered on the coordinate system of the transformed
data. It is important to note that all we have done here is make the eigenvectors
the
 Eigenvalues & Eigenvectors (Con’t)

new coordinate system (y1,y2). Because the covariance matrix of the y's is
diagonal, this in fact also decorrelated the data. The fact that the main data
spread is along e1 is due to the fact that the rows of the transformation matrix A
were chosen according the order of the eigenvalues, with the first row being the
eigenvector corresponding to the largest eigenvalue.
Eigenvalues & Eigenvectors (Con’t)
30-01-2025 Dr. HARIPRASAD S, AP/CINTEL
 Objective

To provide background material in support of topics in Digital Image Processing that

are based on linear system theory.
 Some Definitions

With reference to the following figure, we define a system as a unit that converts an
input function f(x) into an output (or response) function g(x), where x is an independent
variable, such as time or, as in the case of images, spatial position. We assume for
simplicity that x is a continuous variable, but the results that will be derived are equally
applicable to discrete variables.
 Some Definitions (Con’t)

It is required that the system output be determined completely by the input, the
system properties, and a set of initial conditions. From the figure in the previous page,
we write

where H is the system operator, defined as a mapping or assignment of a member of

the set of possible outputs {g(x)} to each member of the set of possible inputs {f(x)}. In
other words, the system operator completely characterizes the system response for a
given set of inputs {f(x)}.
 Some Definitions (Con’t)

An operator H is called a linear operator for a class of inputs {f(x)} if

for all fi(x) and fj(x) belonging to {f(x)}, where the a's are arbitrary constants and

is the output for an arbitrary input fi(x) {f(x)}.

 Some Definitions (Con’t)

The system described by a linear operator is called a linear system (with respect to
the same class of inputs as the operator). The property that performing a linear
process on the sum of inputs is the same that performing the operations individually
and then summing the results is called the property of additivity. The property that
the response of a linear system to a constant times an input is the same as the
response to the original input multiplied by a constant is called the property of
homogeneity.
 Some Definitions (Con’t)

An operator H is called time invariant (if x represents time), spatially invariant (if x is
a spatial variable), or simply fixed parameter, for some class of inputs {f(x)} if

for all fi(x) {f(x)} and for all x0. A system described by a fixed-parameter operator is
said to be a fixed-parameter system. Basically all this means is that offsetting the
independent variable of the input by x0 causes the same offset in the independent
variable of the output. Hence, the input-output relationship remains the same.
 Some Definitions (Con’t)

An operator H is said to be causal, and hence the system described by H is a causal

system, if there is no output before there is an input. In other words,

Finally, a linear system H is said to be stable if its response to any bounded input is
bounded. That is, if

where K and c are constants.

 Some Definitions (Con’t)

Example: Suppose that operator H is the integral operator between the limits − and
x. Then, the output in terms of the input is given by

where w is a dummy variable of integration. This system is linear because

 Some Definitions (Con’t)

We see also that the system is fixed parameter because

where d(w + x0) = dw because x0 is a constant. Following similar manipulation it is

easy to show that this system also is causal and stable.
 Some Definitions (Con’t)

Example: Consider now the system operator whose output is the inverse of the input
so that

In this case,

so this system is not linear. The system, however, is fixed parameter and causal.
 Linear System Characterization-Convolution

A unit impulse, denoted (x − a), is defined by the expression

From the previous sections, the output of a system is given by g(x) = H[f(x)]. But, we
can express f(x) in terms of the impulse just defined, so
 System Characterization (Con’t)

Extending the property of addivity to integrals (recall that an integral can be

approximated by limiting summations) allows us to write

Because f() is independent of x, and using the homogeneity property, it follows

that
 System Characterization (Con’t)

The term

is called the impulse response of H. In other words, h(x, ) is the response of the linear
system to a unit impulse located at coordinate x (the origin of the impulse is the value
of  that produces (0); in this case, this happens when  = x).
 System Characterization (Con’t)

The expression

is called the superposition (or Fredholm) integral of the first kind. This expression is
a fundamental result that is at the core of linear system theory. It states that, if the
response of H to a unit impulse [i.e., h(x, )], is known, then response to any input f
can be computed using the preceding integral. In other words, the response of a
linear system is characterized completely by its impulse response.
 System Characterization (Con’t)

If H is a fixed-parameter operator, then

and the superposition integral becomes

This expression is called the convolution integral. It states that the response of a
linear, fixed-parameter system is completely characterized by the convolution of the
input with the system impulse response. As will be seen shortly, this is a powerful and
most practical result.
 System Characterization (Con’t)

Because the variable  in the preceding equation is integrated out, it is customary to

write the convolution of f and h (both of which are functions of x) as

In other words,
 System Characterization (Con’t)

The Fourier transform of the preceding expression is

The term inside the inner brackets is the Fourier transform of the term h(x −  ). But,
 System Characterization (Con’t)

so,

We have succeeded in proving the important result that the Fourier transform of the
convolution of two functions is the product of their Fourier transforms. As noted
below, this result is the foundation for linear filtering
 System Characterization (Con’t)

Following a similar development, it is not difficult to show that the inverse Fourier
transform of the convolution of H(u) and F(u) [i.e., H(u)*F(u)] is the product f(x)g(x).
This result is known as the convolution theorem, typically written as

and

where "  " is used to indicate that the quantity on the right is obtained by taking the
Fourier transform of the quantity on the left, and, conversely, the quantity on the left
is obtained by taking the inverse Fourier transform of the quantity on the right.
 System Characterization (Con’t)

The mechanics of convolution are explained in detail in the book. We have just filled in
the details of the proof of validity in the preceding paragraphs.

Because the output of our linear, fixed-parameter system is

if we take the Fourier transform of both sides of this expression, it follows from the
convolution theorem that
 System Characterization (Con’t)

The key importance of the result G(u)=H(u)F(u) is that, instead of performing a

convolution to obtain the output of the system, we computer the Fourier transform of
the impulse response and the input, multiply them and then take the inverse Fourier
transform of the product to obtain g(x); that is,

These results are the basis for all the filtering work done in Chapter 4, and some of the
work in Chapter 5 of Digital Image Processing. Those chapters extend the results to
two dimensions, and illustrate their application in considerable detail.
 Objective

To provide background material in support of topics in Digital Image Processing that

are based on probability and random variables.
 Sets and Set Operations

Probability events are modeled as sets, so it is customary to begin a study of

probability by defining sets and some simple operations among sets.

A set is a collection of objects, with each object in a set often referred to as an

element or member of the set. Familiar examples include the set of all image
processing books in the world, the set of prime numbers, and the set of planets
circling the sun. Typically, sets are represented by uppercase letters, such as A,
B, and C, and members of sets by lowercase letters, such as a, b, and c.
 Sets and Set Operations (Con’t)

We denote the fact that an element a belongs to set A by

If a is not an element of A, then we write

A set can be specified by listing all of its elements, or by listing properties

common to all elements. For example, suppose that I is the set of all
integers. A set B consisting the first five nonzero integers is specified using
the notation
 Sets and Set Operations (Con’t)

The set of all integers less than 10 is specified using the notation

which we read as "C is the set of integers such that each members of the set is less
than 10." The "such that" condition is denoted by the symbol “ | “ . As shown in the
previous two equations, the elements of the set are enclosed by curly brackets.

The set with no elements is called the empty or null set, denoted in this review by the
symbol Ø.
 Sets and Set Operations (Con’t)

Two sets A and B are said to be equal if and only if they contain the same
elements. Set equality is denoted by

If the elements of two sets are not the same, we say that the sets are not equal,
and denote this by

If every element of B is also an element of A, we say that B is a subset of A:

 Sets and Set Operations (Con’t)

Finally, we consider the concept of a universal set, which we denote by U and define
to be the set containing all elements of interest in a given situation. For example, in
an experiment of tossing a coin, there are two possible (realistic) outcomes: heads or
tails. If we denote heads by H and tails by T, the universal set in this case is {H,T}.
Similarly, the universal set for the experiment of throwing a single die has six possible
outcomes, which normally are denoted by the face value of the die, so in this case U
= {1,2,3,4,5,6}. For obvious reasons, the universal set is frequently called the sample
space, which we denote by S. It then follows that, for any set A, we assume that Ø 
A  S, and for any element a, a  S and a  Ø.
 Some Basic Set Operations

The operations on sets associated with basic probability theory are straightforward.
The union of two sets A and B, denoted by

is the set of elements that are either in A or in B, or in both. In other words,

Similarly, the intersection of sets A and B, denoted by

is the set of elements common to both A and B; that is,

 Set Operations (Con’t)

Two sets having no elements in common are said to be disjoint or mutually

exclusive, in which case

The complement of set A is defined as

Clearly, (Ac)c=A. Sometimes the complement of A is denoted as .

The difference of two sets A and B, denoted A − B, is the set of elements that
belong to A, but not to B. In other words,
 Set Operations (Con’t)

It is easily verified that

The union operation is applicable to multiple sets. For example the union of sets
A1,A2,…,An is the set of points that belong to at least one of these sets. Similar
comments apply to the intersection of multiple sets.

The following table summarizes several important relationships between sets.

Proofs for these relationships are found in most books dealing with elementary set
theory.
Set Operations (Con’t)
 Set Operations (Con’t)

It often is quite useful to represent sets and sets operations in a so-called Venn
diagram, in which S is represented as a rectangle, sets are represented as areas
(typically circles), and points are associated with elements. The following
example shows various uses of Venn diagrams.

Example: The following figure shows various examples of Venn diagrams. The
shaded areas are the result (sets of points) of the operations indicated in the figure.
The diagrams in the top row are self explanatory. The diagrams in the bottom row
are used to prove the validity of the expression

which is used in the proof of some probability relationships.

Set Operations (Con’t)
 Relative Frequency & Probability

A random experiment is an experiment in which it is not possible to predict

the outcome. Perhaps the best known random experiment is the tossing of a
coin. Assuming that the coin is not biased, we are used to the concept that,
on average, half the tosses will produce heads (H) and the others will produce
tails (T). This is intuitive and we do not question it. In fact, few of us have
taken the time to verify that this is true. If we did, we would make use of the
concept of relative frequency. Let n denote the total number of tosses, nH the
number of heads that turn up, and nT the number of tails. Clearly,
 Relative Frequency & Prob. (Con’t)

Dividing both sides by n gives

The term nH/n is called the relative frequency of the event we have denoted by H,
and similarly for nT/n. If we performed the tossing experiment a large number of
times, we would find that each of these relative frequencies tends toward a stable,
limiting value. We call this value the probability of the event, and denoted it by
P(event).
 Relative Frequency & Prob. (Con’t)

In the current discussion the probabilities of interest are P(H) and P(T). We know in
this case that P(H) = P(T) = 1/2. Note that the event of an experiment need not signify
a single outcome. For example, in the tossing experiment we could let D denote the
event "heads or tails," (note that the event is now a set) and the event E, "neither
heads nor tails." Then, P(D) = 1 and P(E) = 0.

The first important property of P is that, for an event A,

That is, the probability of an event is a positive number bounded by 0 and 1. For
the certain event, S,
 Relative Frequency & Prob. (Con’t)

Here the certain event means that the outcome is from the universal or sample set,
S. Similarly, we have that for the impossible event, Sc

This is the probability of an event being outside the sample set. In the example
given at the end of the previous paragraph, S = D and Sc = E.
 Relative Frequency & Prob. (Con’t)

The event that either events A or B or both have occurred is simply the union of A
and B (recall that events can be sets). Earlier, we denoted the union of two sets by A
 B. One often finds the equivalent notation A+B used interchangeably in
discussions on probability. Similarly, the event that both A and B occurred is given by
the intersection of A and B, which we denoted earlier by A  B. The equivalent
notation AB is used much more frequently to denote the occurrence of both events
in an experiment.
 Relative Frequency & Prob. (Con’t)

Suppose that we conduct our experiment n times. Let n1 be the number of times
that only event A occurs; n2 the number of times that B occurs; n3 the number of
times that AB occurs; and n4 the number of times that neither A nor B occur.
Clearly, n1+n2+n3+n4=n. Using these numbers we obtain the following relative
frequencies:
 Relative Frequency & Prob. (Con’t)

and

Using the previous definition of probability based on relative frequencies we have

the important result

If A and B are mutually exclusive it follows that the set AB is empty and,
consequently, P(AB) = 0.
 Relative Frequency & Prob. (Con’t)

The relative frequency of event A occurring, given that event B has occurred, is given
by

This conditional probability is denoted by P(A/B), where we note the use of the
symbol “ / ” to denote conditional occurrence. It is common terminology to refer
to P(A/B) as the probability of A given B.
 Relative Frequency & Prob. (Con’t)

Similarly, the relative frequency of B occurring, given that A has occurred is

We call this relative frequency the probability of B given A, and denote it by P(B/A).
 Relative Frequency & Prob. (Con’t)

A little manipulation of the preceding results yields the following important

relationships

and

The second expression may be written as

which is known as Bayes' theorem, so named after the 18th century mathematician
Thomas Bayes.
 Relative Frequency & Prob. (Con’t)

Example: Suppose that we want to extend the expression

to three variables, A, B, and C. Recalling that AB is the same as A  B, we replace B

by B  C in the preceding equation to obtain

The second term in the right can be written as

From the Table discussed earlier, we know that

 Relative Frequency & Prob. (Con’t)

so,

Collecting terms gives us the final result

Proceeding in a similar fashion gives

The preceding approach can be used to generalize these expressions to N events.

 Relative Frequency & Prob. (Con’t)

If A and B are statistically independent, then P(B/A) = P(B) and it follows that

and

It was stated earlier that if sets (events) A and B are mutually exclusive, then A  B
= Ø from which it follows that P(AB) = P(A  B) = 0. As was just shown, the two
sets are statistically independent if P(AB)=P(A)P(B), which we assume to be
nonzero in general. Thus, we conclude that for two events to be statistically
independent, they cannot be mutually exclusive.
 Relative Frequency & Prob. (Con’t)

For three events A, B, and C to be independent, it must be true that

and
 Relative Frequency & Prob. (Con’t)

In general, for N events to be statistically independent, it must be true that, for all
combinations 1  i  j  k  . . .  N
 Relative Frequency & Prob. (Con’t)

Example: (a) An experiment consists of throwing a single die twice. The probability
of any of the six faces, 1 through 6, coming up in either experiment is 1/6. Suppose
that we want to find the probability that a 2 comes up, followed by a 4. These two
events are statistically independent (the second event does not depend on the
outcome of the first). Thus, letting A represent a 2 and B a 4,

We would have arrived at the same result by defining "2 followed by 4" to be a
single event, say C. The sample set of all possible outcomes of two throws of a
die is 36. Then, P(C)=1/36.
 Relative Frequency & Prob. (Con’t)

Example (Con’t): (b) Consider now an experiment in which we draw one card from
a standard card deck of 52 cards. Let A denote the event that a king is drawn, B
denote the event that a queen or jack is drawn, and C the event that a diamond-
face card is drawn. A brief review of the previous discussion on relative frequencies
would show that

and
 Relative Frequency & Prob. (Con’t)

Example (Con’t): Furthermore,

and

Events A and B are mutually exclusive (we are drawing only one card, so it would be
impossible to draw a king and a queen or jack simultaneously). Thus, it follows from
the preceding discussion that P(AB) = P(A  B) = 0 [and also that P(AB)  P(A)P(B)].
 Relative Frequency & Prob. (Con’t)

Example (Con’t): (c) As a final experiment, consider the deck of 52 cards again,
and let A1, A2, A3, and A4 represent the events of drawing an ace in each of four
successive draws. If we replace the card drawn before drawing the next card, then
the events are statistically independent and it follows that
 Relative Frequency & Prob. (Con’t)

Example (Con’t): Suppose now that we do not replace the cards that are drawn.
The events then are no longer statistically independent. With reference to the
results in the previous example, we write

Thus we see that not replacing the drawn card reduced our chances of drawing
fours successive aces by a factor of close to 10. This significant difference is perhaps
larger than might be expected from intuition.
 Random Variables

Random variables often are a source of confusion when first encountered. This
need not be so, as the concept of a random variable is in principle quite simple. A
random variable, x, is a real-valued function defined on the events of the sample
space, S. In words, for each event in S, there is a real number that is the
corresponding value of the random variable. Viewed yet another way, a random
variable maps each event in S onto the real line. That is it. A simple, straightforward
definition.
 Random Variables (Con’t)

Part of the confusion often found in connection with random variables is the fact
that they are functions. The notation also is partly responsible for the problem. In
other words, although typically the notation used to denote a random variable is as
we have shown it here, x, or some other appropriate variable, to be strictly formal, a
random variable should be written as a function x(·) where the argument is a
specific event being considered. However, this is seldom done, and, in our
experience, trying to be formal by using function notation complicates the issue
more than the clarity it introduces. Thus, we will opt for the less formal notation,
with the warning that it must be keep clearly in mind that random variables are
functions.
 Random Variables (Con’t)

Example: Consider again the experiment of drawing a single card from a standard
deck of 52 cards. Suppose that we define the following events. A: a heart; B: a
spade; C: a club; and D: a diamond, so that S = {A, B, C, D}. A random variable is
easily defined by letting x = 1 represent event A, x = 2 represent event B, and so on.

As a second illustration, consider the experiment of throwing a single die and

observing the value of the up-face. We can define a random variable as the
numerical outcome of the experiment (i.e., 1 through 6), but there are many other
possibilities. For example, a binary random variable could be defined simply by
letting x = 0 represent the event that the outcome of throw is an even number and
x = 1 otherwise.
 Random Variables (Con’t)

Note the important fact in the examples just given that the probability of the events
have not changed; all a random variable does is map events onto the real line.
 Random Variables (Con’t)

Thus far we have been concerned with random variables whose values are discrete.
To handle continuous random variables we need some additional tools. In the
discrete case, the probabilities of events are numbers between 0 and 1. When
dealing with continuous quantities (which are not denumerable) we can no longer
talk about the "probability of an event" because that probability is zero. This is not
as unfamiliar as it may seem. For example, given a continuous function we know
that the area of the function between two limits a and b is the integral from a to b of
the function. However, the area at a point is zero because the integral from,say, a to
a is zero. We are dealing with the same concept in the case of continuous random
variables.
 Random Variables (Con’t)

Thus, instead of talking about the probability of a specific value, we talk about the
probability that the value of the random variable lies in a specified range. In
particular, we are interested in the probability that the random variable is less than
or equal to (or, similarly, greater than or equal to) a specified constant a. We write
this as

If this function is given for all values of a (i.e., −  < a < ), then the values of random
variable x have been defined. Function F is called the cumulative probability
distribution function or simply the cumulative distribution function (cdf). The
shortened term distribution function also is used.
 Random Variables (Con’t)

Observe that the notation we have used makes no distinction between a random
variable and the values it assumes. If confusion is likely to arise, we can use more
formal notation in which we let capital letters denote the random variable and
lowercase letters denote its values. For example, the cdf using this notation is
written as

When confusion is not likely, the cdf often is written simply as F(x). This notation
will be used in the following discussion when speaking generally about the cdf of a
random variable.
 Random Variables (Con’t)

Due to the fact that it is a probability, the cdf has the following properties:

where x+ = x + , with  being a positive, infinitesimally small number.

 Random Variables (Con’t)

The probability density function (pdf) of random variable x is defined as the derivative
of the cdf:

The term density function is commonly used also. The pdf satisfies the following
properties:
 Random Variables (Con’t)

The preceding concepts are applicable to discrete random variables. In this case,
there is a finite no. of events and we talk about probabilities, rather than
probability density functions. Integrals are replaced by summations and,
sometimes, the random variables are subscripted. For example, in the case of a
discrete variable with N possible values we would denote the probabilities by P(xi),
i=1, 2,…, N.
 Random Variables (Con’t)

In Sec. 3.3 of the book we used the notation p(rk), k = 0,1,…, L - 1, to denote the
histogram of an image with L possible gray levels, rk, k = 0,1,…, L - 1, where p(rk) is the
probability of the kth gray level (random event) occurring. The discrete random
variables in this case are gray levels. It generally is clear from the context whether one is
working with continuous or discrete random variables, and whether the use of
subscripting is necessary for clarity. Also, uppercase letters (e.g., P) are frequently used
to distinguish between probabilities and probability density functions (e.g., p) when
they are used together in the same discussion.
 Random Variables (Con’t)

If a random variable x is transformed by a monotonic transformation function T(x) to

produce a new random variable y, the probability density function of y can be obtained
from knowledge of T(x) and the probability density function of x, as follows:

where the subscripts on the p's are used to denote the fact that they are different
functions, and the vertical bars signify the absolute value. A function T(x) is
monotonically increasing if T(x1) < T(x2) for x1 < x2, and monotonically decreasing if
T(x1) > T(x2) for x1 < x2. The preceding equation is valid if T(x) is an increasing or
decreasing monotonic function.
 Expected Value and Moments

The expected value of a function g(x) of a continuos random variable is defined as

If the random variable is discrete the definition becomes

 Expected Value & Moments (Con’t)

The expected value is one of the operations used most frequently when working with
random variables. For example, the expected value of random variable x is obtained by
letting g(x) = x:

when x is continuos and

when x is discrete. The expected value of x is equal to its average (or mean) value,
hence the use of the equivalent notation and m.
 Expected Value & Moments (Con’t)

The variance of a random variable, denoted by ², is obtained by letting g(x) = x²

which gives

for continuous random variables and

for discrete variables.

 Expected Value & Moments (Con’t)

Of particular importance is the variance of random variables that have been

normalized by subtracting their mean. In this case, the variance is

and

for continuous and discrete random variables, respectively. The square root of the
variance is called the standard deviation, and is denoted by .
 Expected Value & Moments (Con’t)

We can continue along this line of thought and define the nth central moment of a
continuous random variable by letting

and

for discrete variables, where we assume that n  0. Clearly, µ0=1, µ1=0, and µ2=². The
term central when referring to moments indicates that the mean of the random variables
has been subtracted out. The moments defined above in which the mean is not
subtracted out sometimes are called moments about the origin.
 Expected Value & Moments (Con’t)

In image processing, moments are used for a variety of purposes, including histogram
processing, segmentation, and description. In general, moments are used to
characterize the probability density function of a random variable. For example, the
second, third, and fourth central moments are intimately related to the shape of the
probability density function of a random variable. The second central moment (the
centralized variance) is a measure of spread of values of a random variable about its
mean value, the third central moment is a measure of skewness (bias to the left or
right) of the values of x about the mean value, and the fourth moment is a relative
measure of flatness. In general, knowing all the moments of a density specifies that
density.
 Expected Value & Moments (Con’t)

Example: Consider an experiment consisting of repeatedly firing a rifle at a target, and

suppose that we wish to characterize the behavior of bullet impacts on the target in
terms of whether we are shooting high or low.. We divide the target into an upper and
lower region by passing a horizontal line through the bull's-eye. The events of interest
are the vertical distances from the center of an impact hole to the horizontal line just
described. Distances above the line are considered positive and distances below the
line are considered negative. The distance is zero when a bullet hits the line.
 Expected Value & Moments (Con’t)

In this case, we define a random variable directly as the value of the distances in our
sample set. Computing the mean of the random variable indicates whether, on
average, we are shooting high or low. If the mean is zero, we know that the average of
our shots are on the line. However, the mean does not tell us how far our shots
deviated from the horizontal. The variance (or standard deviation) will give us an idea
of the spread of the shots. A small variance indicates a tight grouping (with respect to
the mean, and in the vertical position); a large variance indicates the opposite. Finally,
a third moment of zero would tell us that the spread of the shots is symmetric about
the mean value, a positive third moment would indicate a high bias, and a negative
third moment would tell us that we are shooting low more than we are shooting high
with respect to the mean location.
 The Gaussian Probability Density Function

Because of its importance, we will focus in this tutorial on the Gaussian probability
density function to illustrate many of the preceding concepts, and also as the basis for
generalization to more than one random variable. The reader is referred to Section
5.2.2 of the book for examples of other density functions.

A random variable is called Gaussian if it has a probability density of the form

where m and  are as defined in the previous section. The term normal also is used
to refer to the Gaussian density. A plot and properties of this density function are
given in Section 5.2.2 of the book.
 The Gaussian PDF (Con’t)

The cumulative distribution function corresponding to the Gaussian density is

which, as before, we interpret as the probability that the random variable lies
between minus infinite and an arbitrary value x. This integral has no known closed-
form solution, and it must be solved by numerical or other approximation methods.
Extensive tables exist for the Gaussian cdf.
 Several Random Variables

In the previous example, we used a single random variable to describe the behavior
of rifle shots with respect to a horizontal line passing through the bull's-eye in the
target. Although this is useful information, it certainly leaves a lot to be desired in
terms of telling us how well we are shooting with respect to the center of the target.
In order to do this we need two random variables that will map our events onto the
xy-plane. It is not difficult to see how if we wanted to describe events in 3-D space
we would need three random variables. In general, we consider in this section the
case of n random variables, which we denote by x1, x2,…, xn (the use of n here is not
related to our use of the same symbol to denote the nth moment of a random
variable).
 Several Random Variables (Con’t)

It is convenient to use vector notation when dealing with several random variables.
Thus, we represent a vector random variable x as

Then, for example, the cumulative distribution function introduced earlier

becomes
 Several Random Variables (Con’t)

when using vectors. As before, when confusion is not likely, the cdf of a random
variable vector often is written simply as F(x). This notation will be used in the
following discussion when speaking generally about the cdf of a random variable
vector.

As in the single variable case, the probability density function of a random variable
vector is defined in terms of derivatives of the cdf; that is,
 Several Random Variables (Con’t)

The expected value of a function of x is defined basically as before:

 Several Random Variables (Con’t)

Cases dealing with expectation operations involving pairs of elements of x are

particularly important. For example, the joint moment (about the origin) of order
kq between variables xi and xj
 Several Random Variables (Con’t)

When working with any two random variables (any two elements of x) it is common
practice to simplify the notation by using x and y to denote the random variables. In
this case the joint moment just defined becomes

It is easy to see that k0 is the kth moment of x and 0q is the qth moment of y, as
defined earlier.
 Several Random Variables (Con’t)

The moment 11 = E[xy] is called the correlation of x and y. As discussed in Chapters 4
and 12 of the book, correlation is an important concept in image processing. In fact, it
is important in most areas of signal processing, where typically it is given a special
symbol, such as Rxy:
 Several Random Variables (Con’t)

If the condition

holds, then the two random variables are said to be uncorrelated. From our earlier
discussion, we know that if x and y are statistically independent, then p(x, y) = p(x)p(y),
in which case we write

Thus, we see that if two random variables are statistically independent then they
are also uncorrelated. The converse of this statement is not true in general.
 Several Random Variables (Con’t)

The joint central moment of order kq involving random variables x and y is defined
as

where mx = E[x] and my = E[y] are the means of x and y, as defined earlier. We note
that

and

are the variances of x and y, respectively.

 Several Random Variables (Con’t)

The moment µ11

is called the covariance of x and y. As in the case of correlation, the covariance is an

important concept, usually given a special symbol such as Cxy.
 Several Random Variables (Con’t)

By direct expansion of the terms inside the expected value brackets, and recalling the
mx = E[x] and my = E[y], it is straightforward to show that

From our discussion on correlation, we see that the covariance is zero if the random
variables are either uncorrelated or statistically independent. This is an important
result worth remembering.
 Several Random Variables (Con’t)

If we divide the covariance by the square root of the product of the variances we
obtain

The quantity  is called the correlation coefficient of random variables x and y. It can
be shown that  is in the range −1    1 (see Problem 12.5). As discussed in Section
12.2.1, the correlation coefficient is used in image processing for matching.
 The Multivariate Gaussian Density

As an illustration of a probability density function of more than one random variable,

we consider the multivariate Gaussian probability density function, defined as

where n is the dimensionality (number of components) of the random vector x, C

is the covariance matrix (to be defined below), |C| is the determinant of matrix C,
m is the mean vector (also to be defined below) and T indicates transposition (see
the review of matrices and vectors).
 The Multivariate Gaussian Density (Con’t)

The mean vector is defined as

and the covariance matrix is defined as

 The Multivariate Gaussian Density (Con’t)

The element of C are the covariances of the elements of x, such that

where, for example, xi is the ith component of x and mi is the ith component of m.
 The Multivariate Gaussian Density (Con’t)

Covariance matrices are real and symmetric (see the review of matrices and vectors).
The elements along the main diagonal of C are the variances of the elements x, such
that cii= xi². When all the elements of x are uncorrelated or statistically independent,
cij = 0, and the covariance matrix becomes a diagonal matrix. If all the variances are
equal, then the covariance matrix becomes proportional to the identity matrix, with
the constant of proportionality being the variance of the elements of x.
 The Multivariate Gaussian Density (Con’t)

Example: Consider the following bivariate (n = 2) Gaussian probability density

function

with

and
 The Multivariate Gaussian Density (Con’t)

where, because C is known to be symmetric, c12 = c21. A schematic diagram of this

density is shown in Part (a) of the following figure. Part (b) is a horizontal slice of Part
(a). From the review of vectors and matrices, we know that the main directions of data
spread are in the directions of the eigenvectors of C. Furthermore, if the variables are
uncorrelated or statistically independent, the covariance matrix will be diagonal and the
eigenvectors will be in the same direction as the coordinate axes x1 and x2 (and the
ellipse shown would be oriented along the x1 - and x2-axis). If, the variances along the
main diagonal are equal, the density would be symmetrical in all directions (in the form
of a bell) and Part (b) would be a circle. Note in Parts (a) and (b) that the density is
centered at the mean values (m1,m2).
The Multivariate Gaussian Density (Con’t)
 Linear Transformations of Random Variables

As discussed in the Review of Matrices and Vectors, a linear transformation of a

vector x to produce a vector y is of the form y = Ax. Of particular importance in our
work is the case when the rows of A are the eigenvectors of the covariance matrix.
Because C is real and symmetric, we know from the discussion in the Review of
Matrices and Vectors that it is always possible to find n orthonormal eigenvectors
from which to form A. The implications of this are discussed in considerable detail
at the end of the Review of Matrices and Vectors, which we recommend should be
read again as a conclusion to the present discussion.