You are on page 1of 25

Pattern Recognition and Machine

Learning
Dr Suresh Sundaram
sureshsundaram@iitg.ernet.in

Logistics

Wed :
Thur:
Fri :
Venue:

9 am- 9:55 am
10 am- 10: 55 am
11 am -11:55 am
1207

75% attendance mandatory !


Losing out on classes may leave you out of sync
with the course, leading to low grades !

Pre requisites
Strong foundations in linear algebra,
probability and optimization.
You are warned that if you lack these basics,
you ll have a tough time battling EE 657!

Reference books
Pattern Classification : - Duda, Hart, Stork
Pattern recognition and Machine Learning :Christopher Bishop
Neural networks for Pattern Recognition :Christopher Bishop
Introduction to Machine Learning :- Alpaydin

Reference Books
Machine Learning :- Tom Mitchell
Pattern Recognition :- Sergios Theodoridis
Machine learning : a probabilistic perspective
:- Kevin Murphy

Journals

IEEE TPAMI
Pattern Recognition
Pattern Recognition Letters
Pattern Analysis and Applications
IEEE TIP

IEEE Multimedia
Speech Technology

Conference

ICPR
ICVGIP
ICASSP
NIPS
ICML
ECCV
ACCV.

Grading

Two to three quizzes :- 25 marks


Mid Term
:- 30 marks
End Term
:- 35 marks
Assignments
:- 10 marks

Zero TOLERANCE to copying !

Course Material
A fine blend of black board work with slide shows
The slides will be uploaded as soon as a class is
completed
The slides may only give a glimpse of the course
lecture for better understanding , you are suggested
to strongly read the appropriate sections of the
prescribed books.
Register on moodle

Log in

PRML2015

Lets get started


Person identification systems -> Biometrics,
Aadhar,

Human Perception
How did we learn the alphabet of the English
language?
Trained ourselves to recognize alphabets, so
that given a new alphabet, we use our
memory / intelligence in recognizing it.

Machine Perception
How about providing such capabilities to
machines to recognize alphabets ?
The field of pattern recognition exactly does
that.

Idea
Build a machine that can recognize patterns:
Speech recognition
Fingerprint identification

OCR (Optical Character Recognition)


DNA sequence identification

A basic PR framework
Training samples
Testing samples
An algorithm for recognizing an unknown test
sample
Samples are labeled (supervised learning)

Typical supervised PR problem


Alphabets 26 in number (upper case)
# of alphabets/ classes to recognize 26.
Collect samples of each of the 26 alphabets
and train using an algorithm.
Once trained, test system using unknown test
sample/ alphabeth.

Basics

So what's a pattern ?

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
alphabeth

Handwriting Recognition

Machine print document

Input handwritten document

Handwriting recognition

Face recognition

Fingerprint recognition

Other Applications

Object classification
Signature verification ( genuine vs forgery)
Iris recognition
Writer adaptation
Speaker recognition
Bioinformatics (gene classification)
Communication System Design
Medical Image processing

Pattern Recognition Algorithms


Bag of algorithms that can used to provide
some intelligence to a machine.
These algorithms have a solid probabilistic
framework.
Algorithms work on certain characteristics
defining a class -refered as features.

What is a feature?
Features across classes need to be
discriminative for better classification
peformance.
Pattern

Pattern

Presence of a dot in i can distinguish these


i from l and is a feature.
Features values can be discrete or continuous
in nature (floating value).
In practice, a single feature may not suffice for
discrimination.

You might also like