You are on page 1of 15

Digital Image Processing

Lecturer #9

Dr. Md. Hasanuzzaman


Associate Professor
Department of Computer Science & Engineering
University of Dhaka

11/27/09 Digital Image Processing, Lecture #9 1


Contents
 Pattern Recognition Approaches
 Template Matching
 PCA Method
 Subspace Method
 Knowledge-Based Approach
 Neural Network
 Hidden Markov Model (HMM)

11/27/09 Digital Image Processing, Lecture #9 2


Principal Component Analysis (1 of 10)

 Prepare training images for face or


predefined hand poses, Xi(NxN)
 Calculate eigenvectors (um) from training
images
 Select k-number of eigenvectors (u )
k

corresponding to the highest eigenvalues


to form PCA.
11/27/09 Digital Image Processing, Lecture #9 3
Principal Component Analysis (2 of 10)
Example: Face Recognition

Test

Training Images

11/27/09 Digital Image Processing, Lecture #9 4


Principal Component Analysis (3 of 10)
Example: ASL Characters Recognition

26 ASL Characters (A-R)


11/27/09 Digital Image Processing, Lecture #9 5
Principal Component Analysis (4 of 10)
Example: ASL Characters Recognition

26 ASL Characters (S-Z)


11/27/09 Digital Image Processing, Lecture #9 6
Principal Component Analysis (5 of 10)
Example: PCA Formation

Example of eigenface images

Number of eigenfaces
11/27/09 Digital Image Processing, Lecture #9 7
Principal Component Analysis (6 of 10)

 Known images are projected on k-


eigenimages to form weight vectors:
wk =
u k X−
(i Φ
T
)

=i
[ w,1w,...,
2 k
w]
Φ= Average image, Xi= ith Training image

11/27/09 Digital Image Processing, Lecture #9 8


Principal Component Analysis (7 of 10)
Features Points for 6-ASL Characters

11/27/09 Digital Image Processing, Lecture #9 9


Principal Component Analysis (8 of 10)
Features Points for 6-ASL Characters
Test
¤ point

Result= D

11/27/09 Digital Image Processing, Lecture #9 10


Principal Component Analysis (9 of 10)

 Each test image is projected on the k-eigenimages


and calculated weight vectors (Ω) by
 Classify the pose based on minimum Euclidean
distance.

ε j =Ω
|| −Ωi ||
ε = arg min{ε j }
11/27/09 Digital Image Processing, Lecture #9 11
Principal Component Analysis (10 of 10)

 If minimum Euclidian distance is lower


than the predefined threshold value then
corresponding pose is recognized.
i.e, (ε<Thi), then the image is classified as jth pose

11/27/09 Digital Image Processing, Lecture #9 12


Presentation G1
On 19 April
 Knowledge & Intelligence
 What is AI?
 History of AI

 Knowledge Representation systems


 Rule-based system
 Frame-based System
 Semantic Net.

11/27/09 Digital Image Processing, Lecture #9 13


Presentation G2
On 19 April
 Agent & Intelligent Agents
 The Structure of Agents
 Different Types of Agents
 Simple Reflex Agents
 Model-Based Reflex Agents
 Goal-Based Agents
 Utility Based Agents
 Learning Agents
 Knowledge-Based Agents

11/27/09 Digital Image Processing, Lecture #9 14


Questions?
Or
Suggestions?

Thanks to all

11/27/09 Digital Image Processing, Lecture #9 15

You might also like