CSSE463: Image Recognition Day 27



Similar documents
STATISTICAL PROPERTIES OF LEAST SQUARES ESTIMATORS. x, where. = y - ˆ " 1

ANOVA Notes Page 1. Analysis of Variance for a One-Way Classification of Data

Simple Linear Regression

MDM 4U PRACTICE EXAMINATION

Robust Realtime Face Recognition And Tracking System

Preprocess a planar map S. Given a query point p, report the face of S containing p. Goal: O(n)-size data structure that enables O(log n) query time.

6.7 Network analysis Introduction. References - Network analysis. Topological analysis

Chapter Eight. f : R R

How To Value An Annuity

Average Price Ratios

The simple linear Regression Model

Numerical Methods with MS Excel

Lecture 7. Norms and Condition Numbers

FINANCIAL MATHEMATICS 12 MARCH 2014

IDENTIFICATION OF THE DYNAMICS OF THE GOOGLE S RANKING ALGORITHM. A. Khaki Sedigh, Mehdi Roudaki

Geometric Motion Planning and Formation Optimization for a Fleet of Nonholonomic Wheeled Mobile Robots

OPTIMAL KNOWLEDGE FLOW ON THE INTERNET

10.5 Future Value and Present Value of a General Annuity Due

Abraham Zaks. Technion I.I.T. Haifa ISRAEL. and. University of Haifa, Haifa ISRAEL. Abstract

CIS603 - Artificial Intelligence. Logistic regression. (some material adopted from notes by M. Hauskrecht) CIS603 - AI. Supervised learning

T = 1/freq, T = 2/freq, T = i/freq, T = n (number of cash flows = freq n) are :

CH. V ME256 STATICS Center of Gravity, Centroid, and Moment of Inertia CENTER OF GRAVITY AND CENTROID

Raport końcowy Zadanie nr 8:

STATIC ANALYSIS OF TENSEGRITY STRUCTURES

Chapter 3. AMORTIZATION OF LOAN. SINKING FUNDS R =

n. We know that the sum of squares of p independent standard normal variables has a chi square distribution with p degrees of freedom.

CHAPTER 2. Time Value of Money 6-1

Classic Problems at a Glance using the TVM Solver

1. The Time Value of Money

Key players and activities across the ERP life cycle: A temporal perspective

Relaxation Methods for Iterative Solution to Linear Systems of Equations

Constrained Cubic Spline Interpolation for Chemical Engineering Applications

Settlement Prediction by Spatial-temporal Random Process

Approximation Algorithms for Scheduling with Rejection on Two Unrelated Parallel Machines

THE McELIECE CRYPTOSYSTEM WITH ARRAY CODES. MATRİS KODLAR İLE McELIECE ŞİFRELEME SİSTEMİ

Speeding up k-means Clustering by Bootstrap Averaging

Statistical Pattern Recognition (CE-725) Department of Computer Engineering Sharif University of Technology

Session 4: Descriptive statistics and exporting Stata results

where p is the centroid of the neighbors of p. Consider the eigenvector problem

Chapter = 3000 ( ( 1 ) Present Value of an Annuity. Section 4 Present Value of an Annuity; Amortization

Finite Dimensional Vector Spaces.

Curve Fitting and Solution of Equation

On Error Detection with Block Codes

M. Salahi, F. Mehrdoust, F. Piri. CVaR Robust Mean-CVaR Portfolio Optimization

21 Vectors: The Cross Product & Torque

Credibility Premium Calculation in Motor Third-Party Liability Insurance

AP Statistics 2006 Free-Response Questions Form B

APPENDIX III THE ENVELOPE PROPERTY

The Digital Signature Scheme MQQ-SIG

The analysis of annuities relies on the formula for geometric sums: r k = rn+1 1 r 1. (2.1) k=0

A Covariance Analysis Model for DDoS Attack Detection*

ECONOMIC CHOICE OF OPTIMUM FEEDER CABLE CONSIDERING RISK ANALYSIS. University of Brasilia (UnB) and The Brazilian Regulatory Agency (ANEEL), Brazil

The Time Value of Money

Optimal multi-degree reduction of Bézier curves with constraints of endpoints continuity

Banking (Early Repayment of Housing Loans) Order,

Fractal-Structured Karatsuba`s Algorithm for Binary Field Multiplication: FK

Lecture 13 Time Series: Stationarity, AR(p) & MA(q)

A Parallel Transmission Remote Backup System

Cyber Journals: Multidisciplinary Journals in Science and Technology, Journal of Selected Areas in Telecommunications (JSAT), January Edition, 2011

Load and Resistance Factor Design (LRFD)

Web Services Wind Tunnel: On Performance Testing Large-scale Stateful Web Services

Vibration and Speedy Transportation

16. Mean Square Estimation

Luby s Alg. for Maximal Independent Sets using Pairwise Independence

SHAPIRO-WILK TEST FOR NORMALITY WITH KNOWN MEAN

ON SLANT HELICES AND GENERAL HELICES IN EUCLIDEAN n -SPACE. Yusuf YAYLI 1, Evren ZIPLAR 2. yayli@science.ankara.edu.tr. evrenziplar@yahoo.

A Study of Unrelated Parallel-Machine Scheduling with Deteriorating Maintenance Activities to Minimize the Total Completion Time

Simple Interest Loans (Section 5.1) :

CHAPTER 13. Simple Linear Regression LEARNING OBJECTIVES. USING Sunflowers Apparel

Reinsurance and the distribution of term insurance claims

10/19/2011. Financial Mathematics. Lecture 24 Annuities. Ana NoraEvans 403 Kerchof

Efficient Traceback of DoS Attacks using Small Worlds in MANET

of the relationship between time and the value of money.

Group Nearest Neighbor Queries

An Introduction To Error Propagation: Derivation, Meaning and Examples C Y

Face Verification Problem. Face Recognition Problem. Application: Access Control. Biometric Authentication. Face Verification (1:1 matching)

1. Math 210 Finite Mathematics

Automated Event Registration System in Corporation

Forecasting the Direction and Strength of Stock Market Movement

ENTROPİ OPTİMİZASYON ÖLÇÜSÜ İLE OPTİMAL PORTFÖY SEÇİMİ VE BİST ULUSAL-30 ENDEKSİ ÜZERİNE BİR ÇALIŞMA

Optimal Packetization Interval for VoIP Applications Over IEEE Networks

Three Dimensional Interpolation of Video Signals

Performance Attribution. Methodology Overview

ADAPTATION OF SHAPIRO-WILK TEST TO THE CASE OF KNOWN MEAN

Sequences and Series

The Gompertz-Makeham distribution. Fredrik Norström. Supervisor: Yuri Belyaev

Fast, Secure Encryption for Indexing in a Column-Oriented DBMS

Fundamentals of Mass Transfer

Transcription:

CSSE463: Image Recogto Da 27 Ths week Toda: Alcatos of PCA Suda ght: roject las ad relm work due Questos?

Prcal Comoets Aalss weght grth c ( )( ) ( )( ( )( ) ) heght sze Gve a set of samles, fd the drecto(s) of greatest varace. We ve doe ths! Eamle: Satal momets Prcal aes are egevectors of covarace matr Egevalues gave relatve mortace of each dmeso Note that each ot ca be rereseted 2D usg the ew coordate sstem defed b the egevectors The D reresetato obtaed b rojectg the ot oto the rcal as s a reasoabl-good aromato

Covarace Matr (usg matr oeratos) Place the ots ther ow colum. Fd the mea of each row. Subtract t. Multl N * N T You wll get a 22 matr, whch each etr s a summato over all ots. You could the dvde b c ) )( ( ) )( ( ) )( ( F...... 3 2 3 2 N...... 3 2 3 2 Q

Geerc rocess The covarace matr of a set of data gves the was whch the set vares. The egevectors corresodg to the largest egevalues gve the drectos whch t vares most. Two alcatos Egefaces Tme-elased hotograh

Egefaces Questo: what are the rmar was whch faces var? What haes whe we al PCA? For each face, create a colum vector that cotas the test of all the els from that face Ths s a ot a hgh dmesoal sace (e.g., 65536 for a 256256 el mage) Create a matr F of all M faces the trag set. Subtract off the average face, m, to get N Comute the rc rc covarace matr C = N*N T. F, 2, 3, rc,,2 2,2 3,2 rc,2,3 2,3 3,3 rc,3, M 2, M 3, M rc, M M. Turk ad A. Petlad, Egefaces for Recogto, J Cog Neurosc, 3()

Questo: what are the rmar was whch faces var? What haes whe we al PCA? The egevectors are the drectos of greatest varablt Note that these are 65536-D; thus form a face. Ths s a egeface Here are the frst 4 from the ORL face dataset. Egefaces Q2-3

Questo: what are the rmar was whch faces var? What haes whe we al PCA? The egevectors are the drectos of greatest varablt Note that these are 65536-D; thus form a face. Ths s a egeface Here are the frst 4 from the ORL face dataset. Egefaces htt://uload.wkmeda.org/wkeda/commos/6/67/egefaces.g; from the ORL face database, AT&T Laboratores Cambrdge Q2-3

Iterlude: Projectg ots oto les weght grth sze We ca roject each ot oto the rcal as. How? heght

Iterlude: Projectg a ot oto a le Assumg the as s rereseted b a ut vector u, we ca just take the dot-roduct of the ot ad the vector. u* = u T (whch s D) Eamle: Project (5,2) oto le =. If we wat to roject oto two vectors, u ad v smultaeousl: Create w = [u v], the comute w T, whch s 2D. Result: s ow terms of u ad v. Ths geeralzes to arbtrar dmesos. Q4

Alcato: Face detecto If we wat to roject a ot oto two vectors, u ad v smultaeousl: Create w = [u v], the comute w T, whch s 2D. Result: s ow terms of u ad v. I arbtrar dmesos, stll take the dot roduct wth egevectors! You ca rereset a face terms of ts egefaces; t s just a dfferet bass. The M most mortat egevectors cature most of the varablt: Igore the rest! Istead of 65k dmesos, we ol have M (~50 ractce) Call these 50 dmesos face-sace

Egefaces Questo: what are the rmar was whch faces var? What haes whe we al PCA? Kee ol the to M egefaces for face sace. We ca roject a face oto these egevectors. Thus, a face s a lear combato of the egefaces. Ca classf faces ths lower-d sace. There are comutatoal trcks to make the comutato feasble

Tme-elased hotograh Questo: what are the was that outdoor mages var over tme? Form a matr whch each colum s a mage Fd egs of covarace matr See eamle mages o Dr. B s lato or at the lk below. N Jacobs, N Roma, R Pless, Cosstet Temoral Varatos Ma Outdoor Scees. IEEE Comuter Vso ad Patter Recogto, Meaols, MN, Jue 2007.

Tme-elased hotograh Questo: what are the was that outdoor mages var over tme? The mea ad to 3 egevectors (scaled): Iterretato? N Jacobs, N Roma, R Pless, Cosstet Temoral Varatos Ma Outdoor Scees. IEEE Comuter Vso ad Patter Recogto, Meaols, MN, Jue 2007. Q5-6

Tme-elased hotograh Recall that each mage the dataset s a lear combato of the egemages. mea PC PC2 PC3 = + 492* - 27* +393* = - 2472* + 308* +885* N Jacobs, N Roma, R Pless, Cosstet Temoral Varatos Ma Outdoor Scees. IEEE Comuter Vso ad Patter Recogto, Meaols, MN, Jue 2007.

Tme-elased hotograh Ever mage s rojecto oto the frst egevector N Jacobs, N Roma, R Pless, Cosstet Temoral Varatos Ma Outdoor Scees. IEEE Comuter Vso ad Patter Recogto, Meaols, MN, Jue 2007.

Research dea Doe: Fdg the PCs Usg to detect lattude ad logtude gve mages from camera Yet to do: Classfg mages based o ther rojecto to ths sace, as was doe for egefaces