Welcome to Scribd, the world's digital library. Read, publish, and share books and documents. See more
Download
Standard view
Full view
of .
Save to My Library
Look up keyword
Like this
1Activity
0 of .
Results for:
No results containing your search query
P. 1
Exaggerate Self Quotient Image Model For Face Recognition Enlist Subspace Method

Exaggerate Self Quotient Image Model For Face Recognition Enlist Subspace Method

Ratings: (0)|Views: 53 |Likes:
Published by ijcsis
The attribute of reliance facial recognition technique is frequently impinge on variation of illumination such as shadow and illumination direction changes. In this paper, to enrich the performance of self quotient image model for the elimination of lighting effect for face recognition is recited and dispelled in detail. A Histogram equalization is embrace to enhance the contrast of samples. Then normalize the samples as the result of enhanced SQI. To apply enhanced SQI to face recognition subspace analysis algorithms ( PCA, KPCA and ICA ) are tout to perform subspace analysis on normalized samples. To prosecute experiments on CAS-PEAL face database. The face samples are preprocessed by enhance SQI to meliorate the performance of subspace analysis algorithms. Experimental result canonical the preprocessing of samples could salutary the robust to not only lighting but also facial expressions, masking and occlusion etc. in face recognition domain.
The attribute of reliance facial recognition technique is frequently impinge on variation of illumination such as shadow and illumination direction changes. In this paper, to enrich the performance of self quotient image model for the elimination of lighting effect for face recognition is recited and dispelled in detail. A Histogram equalization is embrace to enhance the contrast of samples. Then normalize the samples as the result of enhanced SQI. To apply enhanced SQI to face recognition subspace analysis algorithms ( PCA, KPCA and ICA ) are tout to perform subspace analysis on normalized samples. To prosecute experiments on CAS-PEAL face database. The face samples are preprocessed by enhance SQI to meliorate the performance of subspace analysis algorithms. Experimental result canonical the preprocessing of samples could salutary the robust to not only lighting but also facial expressions, masking and occlusion etc. in face recognition domain.

More info:

Published by: ijcsis on Jul 07, 2011
Copyright:Attribution Non-commercial

Availability:

Read on Scribd mobile: iPhone, iPad and Android.
download as PDF, TXT or read online from Scribd
See more
See less

07/07/2011

pdf

text

original

 
 
S.Muruganantham
1
,Assistant professor S.T.Hindu college Nagercoil, India -629003Email:smuru_1970@yahoo.co.in 
 Abstract -
 
The attribute of reliance facialrecognition technique is frequently impingeon variation of illumination such as shadowand illumination direction changes. In thispaper, to enrich the performance of self quotient image model for the elimination of lighting effect for face recognition is recitedand dispelled in detail. A Histogramequalization is embrace to enhance thecontrast of samples. Then normalize thesamples as the result of enhanced SQI. Toapply enhanced SQI to face recognitionsubspace analysis algorithms ( PCA, KPCAand ICA ) are tout to perform subspaceanalysis on normalized samples. Toprosecute experiments on CAS-PEAL facedatabase. The face samples are preprocessedby enhance SQI to meliorate theperformance of subspace analysisalgorithms. Experimental result canonicalthe preprocessing of samples could salutarythe robust to not only lighting but also facialexpressions, masking and occlusion etc. inface recognition domain.Keywords: Face recognition,Histogram equalization, Subspace analysis,illumination.I.
 
Introduction
The illumination problem has been believed to be one of most difficulties in facerecognition. The image variation from lightingchanges is more portentous than that fromdivengent personal identities.T.Jebarajan
2
 PrincipalKings College of EngineeringChennai, India - 600105Up to now many prosperous algorithmare intend for face recognition. Principlecomponent analysis ( PCA ) [1] , Independentcomponent analysis (ICA ) [2] , Fisher’s linear discriminant ( FLD ) [3] are three basicalgorithm for subspace analysis in facerecognition.Besides the QI method, much other technique has been proposed for facerecognition under various illuminationconditions in Holocene year. SQI perpetuate QIto perform illumination subtraction without theneed for alignment and no shadow assumptionas QI does.Recently some researchers paying closesolitious to preprocessing of face images totune-up lighting in an image by initiate somemodels, like the quotient image QI [4,5,6] based methods, illumination cone basedmethods [7,8], total variation based methods(TV) [9,10]. Shashua et al. propound thequotient image which is an image ratio between a test image and a linear unification of three images illuminated by non coplanar lights, authentic only on albedo informationand therefore is illumination free.QI method could improve the performance, but it exigeny a large training setand this is not practical in many predicament.Wang et al. proposed a self quotient imagemodel (SQI) [6].SQI is proposed based on basicconception of QI. SQI implements thenormalization by dividing the low frequency part of the prime image, since the lightingeffects are represented as the low frequency
 
Exaggerate Self Quotient Image Model for FaceRecognition Enlist Subspace Method
(IJCSIS) International Journal of Computer Science and Information Security,Vol. 9, No. 6, June 2011264http://sites.google.com/site/ijcsis/ISSN 1947-5500
 
 part. But the edge information is over smoothed by the smoother filter. To overcomethis problem by applying lighting invariant preprocessing methods.In order to vanquish the drawback of SQI, a new method enhanced self quotientimage is proposed and dispeled in detail. Toapply ESQI to face recognition, feature isessensed using subspace method and the performance of PCA, KPCA and ICA areascertained and compared. A large scale CAS-PEAL face database is used for check thevalidity of algorithms. Compared with SQIface sample preprocessed, the ESQI modelcould improve the performance of facerecognition algorithm.
II.
 
SQI frame work 
The Lambertian model can bedecomposition into two parts 1. The intrinsic part. 2. The extrinsic part.I (x, y) = (x, y) n(x, y)
T
· S -------------- (1)= F (x, y) · S
 
Where is albedo and n is the surface normal.The main problem for accurate facerecognition is to separating the two factors anddetach the extrinsic factors are a key toattaining robust face recognition.
III.
 
Self quotient image
We define the self quotient image as anintrinsic property of face image. The self quotient image Q of image I is ipsofacto byQ == ---------------- (2)Where * is convolution operator, I is thesmoothed I and F is the smoothing kernel andthe division id point-wise as in the primequotient image. The self quotient image hasillumination invariant properties as ostensive below using Lambertian model but withshadow. When shadow present the Lambertianmodel equation can be epitomize aI = max ( n
T
· S, 0) ------------- (3)
 Analysis
In a self quotient image, we regardthree regions with divergent shapes andshadow conditions.Region 1:In the absence of shadow and presentssmall surface normal variation. In this case,
n
T
(u,v)
C
1 
, where
C
1
is a constant, then wehaveQ(u,v)=
= ------- (4)In this case Q is quantisly illuminationfree and be contingent on the albedo of the faceRegion 2:In the absence of shadow but presentlarge surface normal variation. In this case
n
T
(u,v) S
is not a constant. The SQI is given byQ(u,v) = = -------- (5)In this case Q is contingent on surface normal,albedo and illuminationRegion 3In shadow regions. In this case the grayvalue of image is relatively low and does notvary. Let us feigned that light is harmoniouslyscattered from all directions in shadow regions.The summation of the dot product between nand S
i
is constant in such region. N
T
(u,v)·S
i
(u,v)=n
T
(u,v)·S
i
(u,v)=C
2
 Where C
2
is a constant and
(IJCSIS) International Journal of Computer Science and Information Security,Vol. 9, No. 6, June 2011265http://sites.google.com/site/ijcsis/ISSN 1947-5500
 
Thus I(u,v) in shadow region can bewritten as I(u,v)
C
2
(u,v), then we haveQ(u,v) = =-----(6)In such region Q is illumination freeand also secluds the shadow effect.The advantage of SQI method is 1.Does not need alignment procedure. 2. Notraining images are exigency. 3. SQI is goodfor eradicate shadow. 4. Lightning source can be any type.
IV. Proposed method
In our proposed method we use aweighted Gaussian filter, given byF = W G --------------------- (7)Where W is the weight and G is the Gaussiankernel.In this application we adopt this filter with a multi scale version. Histogramequalization (HE) method has beencountenanced to be a cogent method to adjustlighting in images. Some Examples of  preprocessed images by HE are illustrated infig -1( row marked with HE)Accessory Distance Expression LightingOriginalHistogram equalization(HE)SQIESQIFig-1From Fig – 1 HE method does discardmost of impact of lighting from face images.But HE has some obstacle. The face image preprocessed by HE method some importantregions are blocked. Face feature like eyes noseand mouth are not persist in these blockedregions. In our proposed method ESQI to persist some important features after HEmethod. Fig -1 also indicates some examples of normalized face samples got by our model(row marked with ESQI). Analogous with HEsome consequential regions are not blocked.Analogous with ancestral HE method and SQImethod these advantages of our proposedmodel make normalized face samples got byour model more apt to face recognition on largescale face database.Given the training set X={x
1
,x
2
,……… x
 N
}where N is the number of samples.In order to enhance contrast of everysample x
i
wield histogram equalization, a flathistogram H
1
with gray level K 
1
is procreatedwhere i = 1,2…………. , N.H
1
= * (n
2
/k 
1
) ---------------- (8)A gray scale transformation T (· ) ischosen to minimize
C
1
(T (K) )
 ─ 
C
0
( K )
 Where C
0
(·) is the cumulative histogram of x
i
.To procure the grayscale transformationT(·) minimization of equation (8) may berewritten to minimize the error intervening theoptimal and actual cumulative histogram.The error is defined by
E
rr 
={C
T
- } + ( ) --- (9)
(IJCSIS) International Journal of Computer Science and Information Security,Vol. 9, No. 6, June 2011266http://sites.google.com/site/ijcsis/ISSN 1947-5500

You're Reading a Free Preview

Download
/*********** DO NOT ALTER ANYTHING BELOW THIS LINE ! ************/ var s_code=s.t();if(s_code)document.write(s_code)//-->