A comparative review of various approaches for feature extraction in Face recognition

A Comparative Review of Various Approaches for Feature Extraction in Face Recognition

Outline1. Introduction2. Topic Discussion3. Literature Review4. Conclusion5. References

2 2

Introduction Biometrics is the science of identifying human beings based on the

measurement and analysis of inherent biological features.

Two biometric Identifiers1. Physiological characteristics2. Behavioural characteristics

Face recognition is a form of biometric identification.

And is used as a secure method of identification for access control mechanisms. 3 3

Introduction(cont’d)

Face Detection

Feature

Extraction

Face Recognition

Identification or verifica

tion

Input Image

Fig 1 : Configuration of a general face recognition structure

A face is a three-dimensional object subject to varying illumination, pose, expression that is to be identified based on its two-dimensional image.

Face recognition is a visual pattern recognition problem.

A face recognition system is expected to identify faces present in images and videos automatically.

4 4

Introduction (cont’d)Facial recognition technology is fast gaining support

across the world

For tackling terrorism Identify frauds in financial institutions For user verification at banks, airports etc Surveillance Human-computer interaction Multimedia management 5 5

Introduction (cont’d)Advantages over other biometric technologies

Natural Non intrusive Easy to use Fast

6 6

Topic Discussion(cont’d)Feature Extraction Approaches

– Appearance Based Approach• Based on the image as two dimensional patterns.• Extract any characteristic from the image that is a feature .

– Geometry Based Approach• Features are extracted using size and relative position of important

components of image. 7 7

Topic Discussion(cont’d)Feature Extraction Approaches(cont’d)

– Template Based Approach• Extract facial feature based on previously designed templates using

appropriate function.

– Colour Based Approach• Uses skin colour to isolate the face area from the non face area in an

image.

8 8

Topic Discussion(cont’d)

9 9

Algorithms for Feature Extraction

1. Principal Component Analysis2. Discrete Cosine Transform3. Linear Discriminant Analysis4. Independent Component Analysis

Topic Discussion(cont’d)1. Principal Component Analysis (PCA)

PCA is a technique that can be used to simplify a dataset It is a way of identifying patterns in data And expressing the data in such a way as to highlight their

similarities and differences. Since patterns in data can be hard to find in data of high

dimension, PCA is a powerful tool for analysing data.10 10

Topic Discussion(cont’d)Principal Component Analysis (cont’d)

Advantages

• PCA can be used for data compression, while ensuring that no

information is lost.

• Low noise sensitivity

• Decreased requirements for capacity and memory

11 11

Topic Discussion(cont’d)Principal Component Analysis (cont’d)

Disadvantages

• The covariance matrix is difficult to be evaluated in an accurate manner

• Even the simplest invariance could not be captured by the PCA unless the

training data explicitly provides this information

• The directions of the maximum variance may be useless for classification

purpose

12 12

Topic Discussion(cont’d)2. Discrete Cosine Transform (DCT)

Face Recognition using DCT involves recognizing the corresponding face image from the database.

The face image obtained from the user is cropped such that only the frontal face image is extracted, eliminating the background.

The image is restricted to a size of 128 × 128 pixels.

13 13

Topic Discussion(cont’d) Discrete Cosine Transform (cont’d)

Pixels exhibit certain level of correlation with neighbouring pixels.

DCT transforms a image from the spatial domain to the frequency domain.

Output array of DCT coefficients contain integers; these can range from -1024 to 1023.

14 14

Topic Discussion(cont’d)Discrete Cosine Transform (cont’d)

Advantages

DCT have the properties of

Decorrelation

Energy compaction

DCT does a better job of concentrating energy in to lower

order coefficients. 15 15

Topic Discussion(cont’d)Discrete Cosine Transform (cont’d)

Disadvantages DCT Features are sensitive to changes in the illumination direction.

Magnitude of the DCT coefficients is not spatially invariant.

16 16

Topic Discussion(cont’d)3. Linear Discriminant Analysis (LDA)

LDA has powerful tools for data reduction and feature extraction.

LDA is a dimensionality reduction technique. It maximize the between - class scattering matrix

measure. Minimize the within – class scatter matrix measure.

17 17

Topic Discussion(cont’d) Linear Discriminant Analysis (cont’d) Advantages

Solve the illumination problem by maximizing the ratio of between-

class scatter to within-class scatter.

LDA based algorithms outperform PCA based ones.

18 18

Topic Discussion(cont’d) Linear Discriminant Analysis (cont’d)

Disadvantages Singularity problem, that is, it fails when all scatter matrices are

singular.

Small Sample Size (SSS) Problem.

19 19

Topic Discussion(cont’d)4. Independent Component Analysis (ICA)

Generalization view of the PCA is known as ICA. Eigenvectors of PCA are replaced by the independent

source vectors in ICA Used to minimize second order and higher order

dependencies in the input. Determines a set of statistically independent variables.

20 20

Topic Discussion(cont’d)Independent Component Analysis (cont’d)

Advantages

– It reconstruct the data better than PCA in the presence of noise.

– Better identifies where the data is concentrated in n-dimensional

space. Disadvantages

– ICA methods show difficulties to handle large number of signals

– ICA does not offer an ordering of the source vectors. 21

Literature Review[1] Fate Bellakdhar, Kais Loukil and Mohamed Abid “Face

recognition approach using Gabor Wavelets, PCA and SVM”2013.

Performance of face recognition system is determined by how to extract feature vector and to classify them.

Gabor representations were used in the algorithms based on global approaches.

PCA approach and SVM is used as a new classifier for pattern recognition. 22 22

Literature Review(cont’d) The performance of the proposed algorithm is tested on the

public and largely used databases of FRGCv2 face and ORL

databases.

This approach consists on combining the magnitude and the

phase of Gabor to extract the characteristic vector.

Here we combined PCA and SVM, and produced better

recognition rate compared to single algorithm. 23 23

Literature Review(cont’d)[2] KiranD.Kadam “Face recognition using Principal Component

Analysis with DCT” 2014.

Here combination PCA and DCT is used to represent accurate face recognition system.

Standard databases such as FACES 94 and ORL are used to test the experimental results

Proves that proposed system achieves more accurate face recognition as compared to individual method. 24

Literature Review(cont’d)

Algorithm Flowchart

Database

Start

Input Image from Database

DCT Preprocessing

Feature Extraction using PCA

Face Matching

End 25 25

Literature Review(cont’d)[3] Priyanka Dhoke, M.P. Parsai“A MATLAB based Face Recognition

using PCA with Back Propagation Neural network” 2014.

Here we use a face recognition with PCA with BPNN. The system consists of a database of a set of facial patterns for each

individual. Characteristic features of PCA called “eigenfaces” are extracted. And then combined with BPNN for subsequent recognition of new

images.

26 26

Literature Review(cont’d)

Eigen Faces

I/p Image

Feature Extraction using PCA

Testing using BPNN

Face recognition results

Face Recognition system using PCA and BPNN27 27

Literature Review(cont’d)S No. Author Title Type of

FeaturesCombination of algorithms

Results

1. Faten Bellakhdhar, Kais Loukil, Mohamed Abid

Face recognition approach using Gabor Wavelets, PCA and SVM.

Eyes, nose and mouth

PCA + SVM Produced better max recognition rate compared to single algorithm.

2. Kiran D Kadam Face recognition using Principal Component Analysis with DCT.

Eyes, nose and mouth

DCT + PCA It achieved the accuracy 99.90% on FACES 94.70% on ORL

• Comparison of Algorithms of feature extraction

28 28

Literature Review (cont’d)S No. Author Title Type of

FeaturesCombination of algorithms

Results

3. Priyanka Dhoke, M.P. Parsai

A MATLAB based Face Recognition using PCA with BPNN

Face features PCA + BPNN It produced fast computation and high accuracy rate. Execution time is only few seconds and acceptance ratio is more than 90%

4. Ajeet Singh, BK Singh, Manish Verma

Comparison of Different Algorithms of Face Recognition

Eyes, nose and mouth

ICA, LDA, SVM

ICA consumes more computation time. SVM has the highest (95.6%) rate of accuracy on ATT database. LDA(86.3%) is ahead of SVM(85.4%) on IFD database

29 29

Conclusion The process of facial recognition involves automated methods to

determine identity, using facial features as essential elements of distinction.

Feature Extraction approaches are Appearace based, Geometry based, template based and colour based.

Feature Extraction algorithm discussed here are PCA, DCT, LDA and ICA.

Face Recognition system is affected differently by different feature extraction algorithm.

30 30

References[1] Faten Bellakdhar, Kais Loukil and Mohamed Abid “Face recognition

approach using Gabor Wavelets, PCA and SVM”2013.[2] KiranD.Kadam “Face recognition using Principal Component Analysis with

DCT” 2014.[3] Priyanka Dhoke, M.P. Parsai“A MATLAB based Face Recognition using PCA

with Back Propagation Neural network” 2014.[4] Ajeet Singh, BK Singh, Manish Verma “Comparison of Different Algorithms

of Face Recognition”2012.[5] Divyarajsnh N. Parmar, Brijesh B. Mehta, “Face Recognition Methods &

Applications”, Jan-Feb 2013.

31 31

32

THANK YOU