International Research Journal of Engineering and Technology (IRJET)
e-ISSN: 2395 -0056
Volume: 04 Issue: 05 | May -2017
p-ISSN: 2395-0072
www.irjet.net
REVIEW OF FACE RECOGNITION TECHNIQUE USING PCA AND BPNN Shalu Kansal1, Assistant Professor Er. Suchitra2 1,2Department
of Computer Science and Engineering
Galaxy Global Group of Institution Dinarpur, Ambala, India ---------------------------------------------------------------------------------------------------------------------------------------------Abstract - Face recognition is one of the interesting research areas in past many years. The reason behind this is its numerous ranges of applications like information security, law enforcement & surveillance, access control, and smart cards etc. It is gaining so much attention in public due to network access via multimedia. Network access with the help of face recognition and verification make recognition system difficult to hack by hackers virtually. It is impossible for them to steal someone password. Face recognition can be done in public security systems by comparing selected face feature of candidate and available face database. This paper focuses on highlighting the strengths and limitations of the earlier proposed classification techniques. The paper provides an insight into the reviewed literature to reveal new aspects of research. Keywords: Biometric, PCA , BPNN, Neural Network.
I.
INTRODUCTION
Biometric is a technique for identification and verification of a person based on their behavioural and physiological characteristics. It is an emerging area. The definition of biometrics comes from Greek word: Bio means life &metry means to measure. There are two types of biometric characteristics: behavioural and physiological. Physiological biometrics includes: face, fingerprint, and hand geometry, retina & iris recognition. Behavioural biometrics includes: signature and voice [1]. A biometric system can works in two ways: verification and identification.  Verification: it is a process of one to one comparison to verify the individual. Its performance is better and faster than identification. II.
PCA
The core of face feature extraction is Eigen face also known as Principal Component Analysis (PCA). PCA represents face data in terms of mean square error (MSE). The samples of face data are easily handled with the help of orthogonal component analysis. PCA is a type of feature reduction method that reduce large face Š 2017, IRJET
|
Impact Factor value: 5.181
|
feature into small face feature indicators that represents the face feature effectively [6]. Let us assume there is N number of face samples. By measuring each face sample, we get K number of indicators, means total no of data is NK. It is process to find small number of face indicators called principal components from large data set. The two main property offer by principal components is that they must be independent of each other, they represents the original face information data as much as possible. There is projection of higher dimensional subspace into lower dimensional subspace with the criteria of minimum reconstruction error. The subspace created by these face feature indicators must be relevant to the largest Eigenvectors obtained by covariance matrix. For classical face feature extraction & facial data representation, PCA is widely used. It is also called dimensionality reduction technique. Starting with L number of face features, to obtain a new sample set Z, we use linear transformation procedure. In this, components of Z are uncorrelated. In next stage, we select the significant components. Existing PCA depended facial recognition systems have high computational cost & memory too. Because of this, it is difficult to scale up. In training stage, all the training data are prepared for calculation of projection matrix, this type of operational mode is called batch mode [2]. It stops when all the training data have been operated. There is a case, if we want to add some training data in existing data; we have to retrain all the collected data. This means that it is difficult to scale up the recognition systems. III.
Linear Discriminant Analysis
Another technique for face feature extraction is Linear Discriminant Analysis (LDA). By finding suitable projection vectors, there is projection of higher dimensional subspace into lower dimensional subspace [3]. The major problem faced by LDA is small sample size. This arises when the number of face samples is smaller than dimension of the face samples which results in computational difficulty. LDA is affected by small sample size problem which is major problem in face recognition which results in low recognition rate of test data. ISO 9001:2008 Certified Journal
|
Page 569