Face recognition process

Preview:

DESCRIPTION

Face recognition process. Plan of the lecture. Face recognition process Most useful tools Principal Components Analysis Support Vector Machines Gabor Wavelets Hough Transform Biometric methods. Face recognition process. Detection. Normalisation. Feature vectors comparison. Feature - PowerPoint PPT Presentation

Citation preview

Face Recognition & Biometric Systems, 2005/2006

Face recognition process

Face Recognition & Biometric Systems, 2005/2006

Plan of the lectureFace recognition processMost useful tools Principal Components Analysis Support Vector Machines Gabor Wavelets Hough TransformBiometric methods

Face Recognition & Biometric Systems, 2005/2006

Face detection: aimsFind a face in the image independent of image size independent of face size for RGB and GS images fast & effective independent from head rotation angleFace location passed to normalisation

Face Recognition & Biometric Systems, 2005/2006

Face detection: toolsGeneralised Hough Transform ellipse detectionSupport Vector Machines (SVM) verificationPCA (back projection) verificationGabor Wavelets feature points detectionColour-based face maps

Face Recognition & Biometric Systems, 2005/2006

Face detection: algorithm

Detection of ”vertical” ellipses face candidatesDetection of ”horizontal” ellipses eye sockets candidatesInitial normalisation and verificationDetection of feature points

Face Recognition & Biometric Systems, 2005/2006

Face tracking

Useful in case of video sequences faster than detection smaller precisionTool: Optical flowTracking of feature points

Face Recognition & Biometric Systems, 2005/2006

Normalisation

Input: image from a camera characteristic points locationTarget: generate an image of invariant

parameters eliminate differences within classes

Face Recognition & Biometric Systems, 2005/2006

Normalisation: tools

Geometrical transformsImage filteringHistogram modifications histogram fitting to a histogram

of the average face imageLighting compensation

Face Recognition & Biometric Systems, 2005/2006

Normalisation: stages

Rotation of non-frontal facesGeometrical normalisationLighting compensationHistogram fitting

Face Recognition & Biometric Systems, 2005/2006

Feature extraction

Input: normalised imageTarget: generate a key which describes the

face algorithm of comparing the keys

Face Recognition & Biometric Systems, 2005/2006

Feature extraction: tools

Principal Component Analysis Linear Discriminant Analysis Local PCA Bayesian MatchingGabor Wavelets

Face Recognition & Biometric Systems, 2005/2006

Feature vectors comparison

Coherent with feature extractionEigenfaces geometric distances SVMDual Eigenfaces image difference classifiedElastic Bunch Graph Matching correlation based

Face Recognition & Biometric Systems, 2005/2006

Multi-method fusion

Many feature extraction methods

S1

S2

Sn

... S

K1

K2

Kn

...

Two images Feature vectors Similarities

K1

K2

Kn

...

Face Recognition & Biometric Systems, 2005/2006

Multi-method fusion

Average similarity weighted meanSVM with polynomial kernelSVM for finding optimal weights

Face Recognition & Biometric Systems, 2005/2006

Tools: PCAApplications: feature extraction – the Eigenfaces

method detection (back projection) Dual Eigenfaces

Stages: training feature extraction feature vectors comparison

Face Recognition & Biometric Systems, 2005/2006

Tools: SVM

Applications: face detection – verification feature vectors comparison detection of lighting direction estimation of head rotation angle multi-method fusion image quality assessment

Face Recognition & Biometric Systems, 2005/2006

Tools: SVMStages: training classification

Main idea: data mapped into higher dimension to

achieve linear separability mapping performed by application of

kernelsProblems with training setParameters must be selected properly

Face Recognition & Biometric Systems, 2005/2006

Tools: Gabor WaveletsApplications: feature extraction (EBGM method) feature points detection face tracking (the detected points are

tracked)Properties: local frequency analysis set of various wavelets prepared comparison: correlation with displacement

estimation

Face Recognition & Biometric Systems, 2005/2006

Tools: GHT

Useful for face detectionProperties: directional image generated (set of

segments) probable ellipse centre for every

segment (based on templates) accumulation of the results for all

the segments in the image

Face Recognition & Biometric Systems, 2005/2006

Biometric methodsTypes of the methods: static dynamic (behavioural)

Requirements: universality distinctiveness permanence collectability performance acceptability circumvention

Face Recognition & Biometric Systems, 2005/2006

Face recognitionAdvantages: low invasiveness high speed identification support systemDrawbacks: relatively low effectiveness changeability of a face face is not always visible

Face Recognition & Biometric Systems, 2005/2006

Fingerprint recognition

Advantages: high effectiveness useful for forensic applicationsDisadvantages: long acquisition time low acceptability

Face Recognition & Biometric Systems, 2005/2006

Iris recognition

Advantages: high distinctiveness universalityDrawbacks: high quality image required low permanence in young age

Face Recognition & Biometric Systems, 2005/2006

Behavioural methods

Gait recognitionVoice recognitionSignature analysis

Face Recognition & Biometric Systems, 2005/2006

Thank you for your attention!

Recommended