Course Code: CSE 337

Course name: Pattern Recognition (Eve)

1
 What is Pattern Recognition?
 A pattern 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
2
 Pattern Recognition Applications
Problem Domain Application Input Pattern Pattern Classes
Document image analysis Optical character Document image Characters, words
recognition
Document classification Internet search Text document Semantic categories
Document classification Junk mail filtering Email Junk/non-junk
Multimedia database Internet search Video clip Video genres
retrieval
Speech recognition Telephone directory Speech waveform Spoken words
assistance
Natural language Information extraction Sentences Parts of speech
processing
Biometric recognition Personal identification Face, iris, fingerprint Authorized users for access
control
Medical Computer aided Microscopic image Cancerous/healthy cell
diagnosis
Military Automatic target Optical or infrared image Target type
recognition
Industrial automation Printed circuit board Intensity or range image Defective/non-defective
inspection product
Industrial automation Fruit sorting Images taken on a conveyor Grade of quality
belt

Remote sensing Forecasting crop yield Multispectral image Land use categories
3
Bioinformatics Sequence analysis DNA sequence Known types of genes
 Pattern Recognition Applications

4 Figure 1: English handwriting recognition.

5 Figure 2: Chinese handwriting recognition.

Figure 3: Arabic handwriting recognition.

6
 Pattern Recognition Applications

Figure 4: Fingerprint recognition.

Figure 5: Cancer detection and grading using microscopic tissue data.

8
 Pattern Recognition Applications

Figure 6: Cancer detection and grading using microscopic tissue data.

9
 Pattern Recognition Applications

Figure 7: Land cover classification using satellite data.

10
 Pattern Recognition Applications

Figure 8: Building and building group recognition using satellite data.

11
 An Example
Problem: Sorting incoming fish on a conveyor belt
according to species.
Assume that we have only two kinds of fish:
 Sea bass,
 Salmon.

Figure 10: (a) Sea bass,(b) Salmon.

12
 An Example: Decision Process
What kind of information can distinguish one species
from the other?
 Length, width, weight, number and shape of fins, tail
shape, etc.
What can cause problems during sensing?
 lighting conditions, position of fish on the conveyor belt,
camera noise, etc.
What are the steps in the process?
 Capture image → isolate fish → take measurements →
make decision.

13
 An Example: Selecting Features
Assume a fisherman told us that a sea bass is generally
longer than a salmon.
We can use length as a feature and decide between sea
bass and salmon according to a threshold on length.
How can we choose this threshold?

14
 An Example: Selecting Features

Figure 11: Histograms of the length feature for two types of fish in training
samples. How can we choose the threshold to make a reliable decision?
15
 An Example: Selecting Features
Even though sea bass is longer than salmon on the
average, there are many examples of fish where this
observation does not hold.
Try another feature: average lightness of the fish
scales.

16
 An Example: Selecting Features

Figure 12: Histograms of the lightness feature for two types of fish in training
samples. It looks easier to choose the threshold but we still cannot make a perfect
decision.
17
 An Example: Multiple Features
Assume we also observed that sea bass are typically
wider than salmon.
We can use two features in our decision:
 lightness:
 width:
Each fish image is now represented as a point (feature
vector)

in a two-dimensional feature space.

18
 An Example: Multiple Features

Figure 13: Scatter plot of lightness and width features for training samples.
We can draw a decision boundary to divide the feature space into two
regions. Does it look better than using only lightness?
19
 Pattern Recognition Systems

Figure 19: Object/process diagram of a pattern recognition system.

20
 The Design Cycle

Figure 20: The design cycle.

Data collection:
 Collecting training and testing data.
 How can we know when we have adequately large and
representative set of samples?

21
 The Design Cycle
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.

22
 The Design Cycle
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?

23
 The Design Cycle
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.

24
 The Design Cycle
Evaluation:
 How can we estimate the performance with training
samples?
 How can we predict the performance with future data?
 Problems of overfitting and generalization.

25
 Summary
Pattern recognition techniques find applications in
many areas: machine learning, statistics, mathematics,
computer science, biology, etc.
There are many sub-problems in the design process.
Many of these problems can indeed be solved.
More complex learning, searching and optimization
algorithms are developed with advances in computer
technology.
There remain many fascinating unsolved problems.

26