IAP Chapter7 2018

Chapter 7:
Introduction to
Object (Pattern) Recognition
Dept. of Telecomm. Eng. DIP2018

Faculty of EEE 1 DHT, HCMUT
7. Pattern Recognition
 A pattern (object) is an entity, vaguely defined, that could
be given a name, e.g.,
– fingerprint image,
– handwritten word,
– human face,
– speech signal,
– DNA sequence,
...
 Pattern recognition is the study of how machines can
– observe the environment,
– learn to distinguish patterns of interest,
– make sound and reasonable decisions about the
categories of the patterns.

7. Pattern Recognition Applications
Example pattern recognition applications:

7. Process of a Pattern Recognition System (1)

 Data acquisition and sensing:
– Measurements of physical variables.
– Important issues: bandwidth, resolution, sensitivity,
distortion, SNR, latency, etc.
 Pre-processing:
– Removal of noise in data.
– Isolation of patterns of interest from the background.
 Feature extraction:
– Finding a new representation in terms of features.
 Model learning and estimation:
– Learning a mapping between features and pattern groups
and categories.
 Classification:
– Using features and learned models to assign a pattern to
a category.
 Post-processing:
– Evaluation of confidence in decisions.
– Exploitation of context to improve performance.
– Combination of experts.

7. Design Cycle (1)

7. Design Cycle (2)
 Data collection:
– Collecting training and testing data.
– How can we know when we have adequately large and
representative set of samples?
 Feature selection:
– Domain dependence and prior information.
– Computational cost and feasibility.
– Discriminative features.
• Similar values for similar patterns.
• Different values for different patterns.
– Invariant features with respect to translation, rotation and
scale.
– Robust features with respect to occlusion, distortion,
deformation, and variations in environment.
7. Design Cycle (3)
 Model selection:
– Domain dependence and prior information.
– Definition of design criteria.
– Parametric vs. non-parametric models.
– Handling of missing features.
– Computational complexity.
– Types of models: templates, decision-theoretic or
statistical, syntactic or structural, neural, and hybrid.
– How can we know how close we are to the true model
underlying the patterns?

7. Design Cycle (4)
 Training:
– How can we learn the rule from data?
– Supervised learning: a teacher provides a category label
or cost for each pattern in the training set.
– Unsupervised learning: the system forms clusters or
natural groupings of the input patterns.
– Reinforcement learning: no desired category is given
but the teacher provides feedback to the system such as
the decision is right or wrong.
 Evaluation:
– How can we estimate the performance with training
samples?
– How can we predict the performance with future data?
– Problems of overfitting and generalization.
7. Pattern Recognition Techniques (1)
 Pattern is an arrangement of descriptors (features).
 Pattern class is a family of patterns that share some
common properties.
 The approaches to pattern recognition developed are
divided into two principal areas: decision-theoretic and
structural
– The first category deals with patterns described using
quantitative descriptors, such as length, area, and
texture.
– The second category deals with patterns best described
by qualitative descriptors, such as the relational
descriptors.


Example: 3 pattern classes (for 3 types of iris flowers)
arranged in vectors with 2 descriptors (2 measurements:
width and length of their petals) for each pattern class.

7. Recognition Based on Decision-Theoretic Methods (1)
 Let x = [x1, x2,…, xn]T for W pattern classes 1, … , W,
di(x) > dj(x), j = 1,…, W, i ≠ j.
 In other words, an unknown pattern x is said to belong to
the ith pattern class if, upon substitution of x into all
decision functions, di(x) yields the largest numerical value.
 Suppose that we define the prototype of each pattern class
to be the mean vector of the patterns of that class:
1
mj 
Nj
 x
x
j
j
 Using the Euclidean distance to determine closeness

reduces the problem to computing the distance measures:
D j (x )  x  m j

 Matching with Minimum Distance Classifier
 The smallest distance is equivalent to evaluating the
functions:
1
d j (x )  xT m j  mTj m j
2
 The decision boundary between classes and for a
minimum distance classifier is
dij (x )  di (x )  d j (x )
1
 xT (mi  m j )  (mi  m j )T (mi  m j )  0
2

 Decision boundary of minimum distance classifier:

 Matching by correlation
We consider it as the basis for finding matches of a sub-
image of size J × K within f(x,y) an image of M × N size ,
where we assume that J ≤ M and K ≤ N
c( x, y )   f ( s, t ) w( x  s, y  t )
s t
for x = 0,1, 2,…, M-1, y = 0,1, 2,…, N-1


 The correlation function has the disadvantage of being
sensitive to changes in the amplitude of f and w.
 For example, doubling all values of f doubles the value of
c(x, y).
 An approach frequently used to overcome this difficulty is
to perform matching via the correlation coefficient
 [ f (s, t )  f (s, t )][w( x  s, y  t )  w]
 ( x, y )  s t
1
 2
 [ f ( s, t )  f ( s, t )]  [ w( x  s, y  t )  w] 
2
2
 s t s t 
 The correlation coefficient is scaled in the range -1 to 1,
independent of scale changes in the amplitude of f and w.

 Optimum Statistical Classifiers
 The probability that a particular pattern x comes from
class i is denoted p(i/x)
 If the pattern classifier decides that x came from j
when it actually came from i, it incurs a loss, denoted
Lij W
rj (x)   Lkj p (k x)
k 1
 From basic probability theory, we know that

p ( A B )   p ( A) p ( B A)  p ( B )
1 W
rj ( x)  
p(x) k 1
Lkj p(x k ) P(k )

 Since 1/p(x) is positive and common to allthe rj(x), j = 1, 2,
…, W, it can be dropped without affecting the relative order
of these functions from the smallest to the largest value.
W
rj (x)   Lkj p ( x k ) P(k )
k 1
 Thus the Bayes classifier assigns an unknown pattern x to

class i if
W W
L
k 1
ki p(x k ) P(k )   Lqj p(x q ) P(q )
q 1
Lij  1   ij
W
rj (x)   (1   kj ) p( x k ) P(k )
k 1
 p(x)  p(x  j ) P( j )

 The Bayes classifier then assigns a pattern x to class i if,
p(x)  p(x i ) P(i )  p(x)  p(x  j ) P( j )
or, equivalently, if
p(x i ) P(i )  p(x  j ) P( j ), j  1,2....,W ;i  j
or in the form of decision functions:
d j (x )  p(x  j ) P( j ), j  1,2,...,W

 Bayes classifier for Gaussian pattern classes
Let us consider a 1-D problem (n = 1) involving two pattern
classes (W = 2) governed by Gaussian densities
d j ( x)  p( x  j ) P( j )
( x  m j )2

1 2 2j
 e P( j )
2 j
j  1, 2


In the n-dimensional case, the Gaussian density of the
vectors in the j th pattern class has the form:
1
1  ( x m j )T Cj 1 ( x m j )
p(x  j )  12
e 2
(2 )n 2 C j
m j  E j {x}
C j  E j {( x  m)( x  m)T }
Approximating:
1
mj 
Nj
 x
x j
1
Cj 
Nj
 xx
x j
T
 m j mT
j

• Bayes decision function for class j is
d j (x )  ln[ p(x /  j ) P( j )]
d j ( x )  ln P( j )  x T C1m j  1 / 2mTj C1m j


 Artificial Neuron Networks
– Ideas stem from the operation of human neural
networks.
– Networks of interconnected nonlinear computing
elements called neurons.

• Perceptron:
w and x are n-dimensional column vectors and wTx is the dot

product of the two vectors. We refer to w as a weight vector and,
as above, to wn+1 as a bias. Giving any pattern vector x from a
vector population, find a set of weights with the property:

For the classes are linearly separable:
The perceptron algorithm for finding (or training) w is
simple as following
Let α > 0 denote a correction increment (also called the

learning increment or the learning rate),
let w(1) be a vector with arbitrary values, and
let wn+1(1) be an arbitrary constant.
Then, do the following for k = 2, 3,…: For a pattern
vector, x(k), at step k,


The notation in previous equations can be simplified if we
add a 1 at the end of every pattern vector and include the
bias in the weight vector. That is, we define
x = [x1, x2, …, xn, 1]T
and
w= [w1, w2,…, wn, wn+1]T
Then,

The perceptron algorithm can be modified as: For any
pattern vector, x(k), at step k

For nonseparable pattern classes:
Let r denote the response we want the perceptron to have for
any pattern during training. The output of the perceptron r is
either +1 or −1. We want to find the augmented weight
vector, w, that minimizes the mean squared error (MSE)
between the desired and actual responses of the perceptron.
The perceptron algorithm for finding w is based on the
least-mean-squared-error (LMSE) algorithm as
A typical range for α is 0.1 < α < 1.0, w(1) is arbitrary.

• Artificial Neuron:
Neural networks are interconnected perceptron-like computing
elements called artificial neurons. These neurons perform the
same computations as the perceptron, but with different
activation function.
Activation function:


• Fully Connected
Neural Network

The outputs of the layer 1 are the components of input vector
x, n = n1 is the dimensionality of x:
The computation performed by neuron i in layer l is given by
for i = 1, 2,…, n and l = 2,…, L. Quantity zi(l) is called the

net (or total) input to neuron i in layer l. The reason for this
terminology is that zi(l) is formed using all outputs from layer
(l −1).
The output (activation value) of neuron i in layer l is given by

Matrix form: Implementation of previous equations by
using matrix operations giving computationally faster.
The number of outputs in layer 1 is always of the same
dimension as an input pattern, x, so its matrix (vector)
form is simple:
The matrix, W(l), that contains all the weights in layer l,

each rows contains the weights for one of the nodes in
layer l:

Then, we can obtain all the sum-of-products computations,
zi(l), for layer l simultaneously:
where a(l−1) is a column vector of dimension nl−1×1

containing the outputs of layer l−1, b(l) is a column vector
of dimension nl×1 containing the bias values of all the
neurons in layer l, and z(l) is an n×1 column vector
containing the net input values, zi(l), i=1, 2,…nl , to all the
nodes in layer l.

Because the activation function is applied to each net input
independently of the others, the outputs of the network at
any layer can be expressed in vector form as:


7. Recognition Based on Structural Methods (1)
 Matching Shape Numbers
 All shapes of order 4, 6, and 8
Order6
Order4
Chain code: 0321 Chain code: 003221

Difference : 3333 Difference : 303303
Shape no. : 3333 Shape no. : 033033
Order8
Chain code: 00332211 Chain code:03032211 Chain code: 00032221

Difference : 30303030 Difference :33133030 Difference : 30033003
Shape no. : 03030303 Shape no. :03033133 Shape no. : 00330033

 Advantages:
1. Matching Shape Numbers suits the processing
structure simple graph, specially becomes by
the line combination.
2. Can solve rotation the question.
3. Matching Shape Numbers most emphatically to the
graph outline, shape similarity also may completely
overcome.
4. The Displacement question definitely may
overcome, because of this method emphatically to the
relative position but is not to the position.

 Disadvantages:
1. It can not uses for a hollow structure.
2. Scaling is a shortcoming which needs to change,
perhaps coordinates the alternative means .
3. Intensity.
4. Mirror problem.
5. The color is unable to recognize.

String Matching
 Suppose that two region boundaries, a and b, are coded
into strings denoted a1, a2, …, an and b1, b2, …, bm,
respectively.
 Let  represent the number of matches between the
two strings, where a match occurs in the k th position if
ak = bk
  max( a , b )  

 A simple measure of similarity between a and b is the
ratio:
 
R 
 max( a , b )  
Hence R is infinite for a perfect match and 0 when none

of the corresponding symbols in and match (in this case)



IAP Chapter7 2018

Uploaded by

Copyright:

Available Formats

IAP Chapter7 2018

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

IAP Chapter7 2018

Uploaded by

Copyright:

Available Formats

Chapter 7:

Dept. of Telecomm. Eng. DIP2018

Dept. of Telecomm. Eng. DIP2018

Dept. of Telecomm. Eng. DIP2018

Dept. of Telecomm. Eng. DIP2018

Dept. of Telecomm. Eng. DIP2018

Dept. of Telecomm. Eng. DIP2018

Dept. of Telecomm. Eng. DIP2018

Dept. of Telecomm. Eng. DIP2018

Dept. of Telecomm. Eng. DIP2018

Dept. of Telecomm. Eng. DIP2018

 Using the Euclidean distance to determine closeness

Dept. of Telecomm. Eng. DIP2018

Dept. of Telecomm. Eng. DIP2018

Dept. of Telecomm. Eng. DIP2018

for x = 0,1, 2,…, M-1, y = 0,1, 2,…, N-1

Dept. of Telecomm. Eng. DIP2018

Dept. of Telecomm. Eng. DIP2018

Dept. of Telecomm. Eng. DIP2018

 From basic probability theory, we know that

Dept. of Telecomm. Eng. DIP2018

 Thus the Bayes classifier assigns an unknown pattern x to

 p(x)  p(x  j ) P( j )

Dept. of Telecomm. Eng. DIP2018

Dept. of Telecomm. Eng. DIP2018

Dept. of Telecomm. Eng. DIP2018

Dept. of Telecomm. Eng. DIP2018

d j ( x )  ln P( j )  x T C1m j  1 / 2mTj C1m j

Dept. of Telecomm. Eng. DIP2018

Dept. of Telecomm. Eng. DIP2018

Dept. of Telecomm. Eng. DIP2018

w and x are n-dimensional column vectors and wTx is the dot

Dept. of Telecomm. Eng. DIP2018

Let α > 0 denote a correction increment (also called the

Dept. of Telecomm. Eng. DIP2018

Dept. of Telecomm. Eng. DIP2018

Dept. of Telecomm. Eng. DIP2018

Dept. of Telecomm. Eng. DIP2018

A typical range for α is 0.1 < α < 1.0, w(1) is arbitrary.

Dept. of Telecomm. Eng. DIP2018

Dept. of Telecomm. Eng. DIP2018

Dept. of Telecomm. Eng. DIP2018

Dept. of Telecomm. Eng. DIP2018

The computation performed by neuron i in layer l is given by

for i = 1, 2,…, n and l = 2,…, L. Quantity zi(l) is called the

Dept. of Telecomm. Eng. DIP2018

The matrix, W(l), that contains all the weights in layer l,

Dept. of Telecomm. Eng. DIP2018

where a(l−1) is a column vector of dimension nl−1×1

Dept. of Telecomm. Eng. DIP2018

Dept. of Telecomm. Eng. DIP2018

Dept. of Telecomm. Eng. DIP2018

Chain code: 0321 Chain code: 003221

Chain code: 00332211 Chain code:03032211 Chain code: 00032221

Dept. of Telecomm. Eng. DIP2018

Dept. of Telecomm. Eng. DIP2018

Dept. of Telecomm. Eng. DIP2018

Dept. of Telecomm. Eng. DIP2018

Hence R is infinite for a perfect match and 0 when none