CSE803 Fall 2014 1 Pattern Recognition Concepts Chapter 4: Shapiro and Stockman How should objects be represented? Algorithms for recognition/matching.

CSE803 Fall 2014 1

Pattern Recognition Concepts Chapter 4: Shapiro and Stockman How should objects be represented? Algorithms for recognition/matching * nearest neighbors * decision tree * decision functions * artificial neural networks How should learning/training be done?

CSE803 Fall 2014 2

Feature Vector Representation

X=[x1, x2, … , xn], each xj a real number

Xj may be object measurement

Xj may be count of object parts

Example: object rep. [#holes, Area, moments, ]

CSE803 Fall 2014 3

Possible features for char rec.

CSE803 Fall 2014 4

Some Terminology

Classes: set of m known classes of objects (a) might have known description for each (b) might have set of samples for each Reject Class: a generic class for objects not in any of the designated known classes Classifier: Assigns object to a class based on features

CSE803 Fall 2014 5

Classification paradigms

CSE803 Fall 2014 6

Discriminant functions

Functions f(x, K) perform some computation on feature vector x

Knowledge K from training or programming is used

Final stage determines class

CSE803 Fall 2014 7

Decision-Tree Classifier Uses subsets of

features in seq. Feature

extraction may be interleaved with classification decisions

Can be easy to design and efficient in execution

CSE803 Fall 2014 8

Decision Trees

#holes

moment ofinertia

#strokes #strokes

best axisdirection

#strokes

- / 1 x w 0 A 8 B

01

2

< t t

2 4

0 1

060

90

0 1

CSE803 Fall 2014 9

Classification using nearest class mean

Compute the Euclidean distance between feature vector X and the mean of each class.

Choose closest class, if close enough (reject otherwise)

Low error rate at left

CSE803 Fall 2014 10

Nearest mean might yield poor results with complex structure

Class 2 has two modes

If modes are detected, two subclass mean vectors can be used

CSE803 Fall 2014 11

Scaling coordinates by std dev

CSE803 Fall 2014 12

Another problem for nearest mean classification If unscaled, object

X is equidistant from each class mean

With scaling X closer to left distribution

Coordinate axes not natural for this data

1D discrimination possible with PCA

CSE803 Fall 2014 13

Receiver Operating Curve ROC

Plots correct detection rate versus false alarm rate

Generally, false alarms go up with attempts to detect higher percentages of known objects

CSE803 Fall 2014 14

Confusion matrix shows empirical performance

CSE803 Fall 2014 15

Bayesian decision-making

CSE803 Fall 2014 16

Normal distribution 0 mean and unit

std deviation Table enables us

to fit histograms and represent them simply

New observation of variable x can then be translated into probability

CSE803 Fall 2014 17

Cherry with bruise Intensities at about 750 nanometers

wavelength Some overlap caused by cherry surface

turning away

CSE803 Fall 2014 18

Parametric models