Cs 900 seminar presentation

CS 900 - Graduate Seminar(Spring / Summer – 2015)

Pattern Recognition

- Baabu Aravind Vellaian Selvarajan

200339484

Introduction

• Pattern Recognition is a branch of ArtificialIntelligence (Machine Learning) [1]

• PR is an area of AI deals with recognition ofpatterns and regularities in data to solveproblems using computable machines

AIPR

AI

2

Human Perception

• Humans have developed highly sophisticatedskills for sensing their environment and takingaccording to their observation [2]

• E.g. Recognizing a face, Understanding Spokenlanguage, Reading Handwriting, Smell of food

3

Machine Perception

• The capability of machines to interpret data ina manner that is similar to the way humanuses their senses to relate the world around [3]

• Simply we can say “Building a machine thatcan recognizing patterns”

4

Machine Leaning

• What is machine learning ?

Machine learning is the science of gettingcomputers to act without being explicitlyprogrammed [4].

5

What is Pattern ?

• A set of features of individual objects

• It is an abstraction, represented by a set ofmeasurements describing a “physical” object

• E.g. Visual, Temporal, Musical, Logical.. Etc.,

6

Pattern Class

• A set of patterns sharing common attributes [5]

7

What is meant by Recognition ?

• Discover to which class of entities the “pattern”belongs and the name of the “pattern”

• Also its different from “identification” [6]

• For Example: Security system searching database fora person

• finding similar one isface identification• searching several picsof a particular personand allowing him is facerecognition

8

Pattern Recognition

It is the study of how machine can

Perceive + Process + Prediction [2]

• Perceive : Interaction with the real-world (i.e.,observing the environment)

• Process : Learn to distinguish patterns ofinterest from their background

• Predication : Making reasonable decisionsabout the categories of patterns

9

Pattern Recognition

Two phase process

Leaning / Training and Detecting / Classifying

Learning:– its time consuming and hard process

– Several examples of each class must be exposed to thesystem

10

Classification Algorithm

It is otherwise called as supervised learning

A teacher provides a category label to train a classifier [2]

11

Clustering Algorithm

It is otherwise called as unsupervised learning

System forms clusters or natural groupings of input patterns based on some similar criteria [2]

12

Pattern Recognition System

https://www.projectrhea.org/rhea/index.php/Introduction_To_Pattern_Recognition_and_Classification

13

Pattern Recognition System

• Sensing – which collects data, the measurementof physical variables

• Segmentation – Isolation of pattern of interestfrom background and removal of noise from thedata

• Feature Extraction – in terms of features findinga new representation

• Classification – using features assign the input tothe category or class

• Post-processing – making decision using thefeatures and classification

14

Applications

Optical Character Recognition

Hand Written: sorting letters, input device for PDA’sPrinted Texts : digitalization of text documents and reading machines for blind people

Biometrics Face Recognition, Verification, RetrievalFinger Print RecognitionSpeech Recognition

Diagnostic systems Medical Diagnosis: X-Ray, Electro Cardio Graph analysis

Military applications

Automated Target RecognitionImage segmentation and analysis – recognition from aerial or satellite photographs

15

Approach

• Statistical Model : Pattern recognition systemsare based on statistics and probabilities

• Syntactic Model / Structural Model: Based onrelation between features, patterns arerepresented by structures

16

Approach

• Template matching model: a template or aprototype of the pattern to be recognized isavailable

• Neural Network Model: able to learn andresolve complex problems based on availableknowledge.

17

Case Study

• Source

–Pattern Classification – 2nd Edition Bookby Richard Duda and Peter Hart

• Problem

–A fish packing plant wants to automate theprocess of sorting incoming fish on aconveyor according to species using opticalsensing

18

Case Study

• Fish Classification

–Considering only two types of fishes

– SeaBass / Salmon

• Camera has been set up for sensing – taking pictures of the incoming fish

19

Case Study

• What can cause problems during sensing ?– Lighting conditions

– Position of fish on conveyor belt

– Camera noise, etc.,

• What are the steps in process ?– Capture image

– Isolate fish

– Take measurements

– Make Decisions

20

Case Study

• What kind of information can distinguish one species for the other ?– Length

– Lightness

– Width

– Number and shape of fins

– Position of the mouth, Etc.,

• Additional info from a fisherman “SeaBass” is generally longer than a “Salmon”

21

Case Study

•Preprocess raw data from camera•Segment isolated fish•Extract features from each fish

- Length, width, brightness, etc.,•Classify Each fish

22

Case Study

23

Conclusion

• What happens if a customer finds “Sea Bass”in there “Salmon” can ? (unhappy, costly price)

• We Should also consider cost of differenterrors we make in our decisions

24

References

[1]. https://en.wikipedia.org/wiki/Pattern_recognition

[2]. http://www.slideshare.net/lgustavomartins/introduction-to-pattern-recognition

[3]. https://en.wikipedia.org/wiki/Machine_perception

[4]. https://www.coursera.org/course/ml

[5]. http://www.slideshare.net/MaazHasan/pattern-recognition-37839488?qid=bf185e66-d1da-421f-8325-931165941321&v=qf1&b=&from_search=30

[6]. http://www.slideshare.net/Randa_Elanwar/what-is-patternrecognition-lecture-1

[7].https://www.projectrhea.org/rhea/index.php/Introduction_To_Pattern_Recognition_and_Classification

[8]. http://homepage.tudelft.nl/a9p19/papers/4PR_Approaches.pdf

25

Thanks for your patience