![International Journal of Computer Science, Engineering and Information Technology (IJCSEIT), Vol.2, No.1, February 2012
DOI : 10.5121/ijcseit.2012.2107 67
FACE RECOGNITION USING DIFFERENT LOCAL
FEATURES WITH DIFFERENT DISTANCE
TECHNIQUES
M.Koteswara Rao1
, K.Veera Swamy2
, K.Anithasheela3
and B.Chandra Mohan4
1
Associate Professor, QIS College of Engineering & Technology, Ongole, A.P, India,
koteshproject@gmail.com
2
Professor, QIS College of Engineering & Technology, Ongole, A.P, India,
kilarivs@yahoo.com
3
Associate Professor, JNTUH, Hyderabad, A.P, India,
kanithasheela@gmail.com
4
Professor, Bapatla Engineering College, A.P, India,
chandrabhuma@gmail.com
ABSTRACT
A face recognition system using different local features with different distance measures is proposed in this
paper. Proposed method is fast and gives accurate detection. Feature vector is based on Eigen values,
Eigen vectors, and diagonal vectors of sub images. Images are partitioned into sub images to detect local
features. Sub partitions are rearranged into vertically and horizontally matrices. Eigen values, Eigenvector
and diagonal vectors are computed for these matrices. Global feature vector is generated for face
recognition. Experiments are performed on benchmark face YALE database. Results indicate that the
proposed method gives better recognition performance in terms of average recognized rate and retrieval
time compared to the existing methods.
KEYWORDS
Vertical centered, Horizontal centered, Eigen vector, Eigen values and distance methods.
1. INTRODUCTION
Face recognition [1] is a computer application for identifying and verifying a person from the face
database. The face recognition system is generally used in security purpose and can be compared
to other techniques such as fingerprint, iris and signature. More face recognition methods are
identify face extract feature, or from an image of the database. These methods may analyze the
relative size, shape, position of the database images. These extracted features [5] are used to
search for other relevant images. So far face recognition methods developed can be classified as
holistic method or local feature method. First method is appearance based technique, which
analyze the distribution of individual faces in face space for holistic features. It can be done by
using global and local features[4]. Concentration on dynamic link matching or graph matching is
considered in local feature [1] method. In global feature method or holistic method concentration
on eigenfaces or similar appearance such as Principal Component Analysis (PCA) is considered.
PCA approach is mainly concentrated on dimensionality reduction. This scheme is based on
linearly projecting the image space to a lower dimensionality space that is also known as Eigen
space. Second method is feature based technique [5], which is concentrated on dimensionality of
input image as well as images in face database. In face recognition system dimensionality
reduction [7] is an essential technique. Block diagram of face recognition system is shown in
Figure 1.](https://image.slidesharecdn.com/2112ijcseit07-180824090734/85/FACE-RECOGNITION-USING-DIFFERENT-LOCAL-FEATURES-WITH-DIFFERENT-DISTANCE-TECHNIQUES-1-320.jpg)
![International Journal of Computer Science, Engineering and Information Technology (IJCSEIT), Vol.2, No.1, February 2012
68
Figure 1. Block diagram of face recognition.
2. DISTANCE MEASURES
Several distance measures are useful for face recognition. Manhattan distance, weighted
angle based Distance, and Minkowski Distance are considered in this work.
2.1. Manhattan Distance
Manhattan distance, also known as the L1-distance, between two points in an Euclidean
space fixed Cartesian coordinate system sum of the lengths of the projections of the line
segment between the points onto the coordinate axes. Notice that the Manhattan distance
depends on the choice on the rotation of the coordinate system, but does not depend on
the translation of the coordinate system or its reflection with respect to a coordinate axis.
Manhattan distance is also known as city block distance.
∑ =
−== =
n
i iip yxZYLZYd 1
||),(),( 1
Where n is the number of variables, and Xi and Yi are the values of the ith variable, at
points Y and Z respectively.
2.2. Weighted angle Distance
Cosine similarity [4] is a measure of similarity between two vectors by measuring the cosine of
the angle between them. The cosine of 0 is 1, and less than 1 for any other angle; the lowest value
of the cosine is -1. The cosine of the angle between two vectors thus determines whether two
vectors are pointing in roughly the same direction. Cosine of two vector s can be written as
Similarity = ∑
∑∑
∑
==
=
==
N
I
I
N
I
I
N
I
II
II
YQ
YQ
YQ
YQ
YQ
1
2
1
2
1
)(*)(
*
*
||||.||||
.
)cos(θ
The resulting similarity ranges from −1 meaning exactly opposite, to 1 meaning exactly the same,
with 0 usually indicating independence, and in-between values indicating intermediate similarity
or dissimilarity. For text matching, the attribute vectors A and B are usually the term frequency](https://image.slidesharecdn.com/2112ijcseit07-180824090734/85/FACE-RECOGNITION-USING-DIFFERENT-LOCAL-FEATURES-WITH-DIFFERENT-DISTANCE-TECHNIQUES-2-320.jpg)
![International Journal of Computer Science, Engineering and Information Technology (IJCSEIT), Vol.2, No.1, February 2012
69
vectors of the documents. The cosine similarity can be seen as a method of normalizing document
length during comparison.
2.3. Minkowski Distance
Minkowski distance is concentrated on Euclidean [4] space, which can be considered as a
generalization of both Euclidean and Manhattan distance for getting more recognition efficiency.
Minkowski distance is based on factor p.
),.....,( 21 nxxxp =
Minkowski diastance is typically used with p being 1 or 2. In the limiting case of p reaching
infinity we obtain the chebyshev distance.
∑=
−==
n
i
pip
iip yxZYLZYd 1
/
)||(),(),(
Minkowski distance is often used when variables are measured on ration scales with absolute zero
value. Variables with a wider range can overpower the result.
Figure 2 Minkowski distance
3. PROPOSED ALGORITHM
Proposed algorithm to get feature vector for the database is given below:
1. Each image is partitioned into sub images. D= r1*r2/k. r1&r2 are the image rows &
columns. K is the equally sized.
2. Convert the each sub image into column data matrix. Each of them can be expressed in the
order of a D-by-N. Ci = {ci1+ci2+ci3+…ciN} with i = 1, 2,…….. K. here N is the total
number of images
3. Calculate mean value for each sub image.
4. Subtract the mean value from column data matrix of each sub image then obtain vertically
centered column data matrix Cvi = {ĉi1+ĉi2+ĉi3+…ĉiN} with i = 1, 2,……...K.
5. Rearrange the elements to get square matrix.
6. Collect Eigen values, Eigen vectors, and diagonal values of the square matrix PVi = {Pi1+ Pi2
+Pi3+……+ PiL} with i = 1, 2….S. here L is the feature of the sub image. Then obtain
training data base matrix Gvi = PVi
T
Cvi = {Gi1+ Gi2+……+ GiN} = 1, 2…K.
7. Repeat the same procedure row data matrix.
8. Reduce the feature size as per the requirement. GVj = (G1j
T
,G2J
T
,……GSj
T
)T
, j = 1,2,..N.
9. Convert the feature into global feature by converting into a single value.](https://image.slidesharecdn.com/2112ijcseit07-180824090734/85/FACE-RECOGNITION-USING-DIFFERENT-LOCAL-FEATURES-WITH-DIFFERENT-DISTANCE-TECHNIQUES-3-320.jpg)
![International Journal of Computer Science, Engineering and Information Technology (IJCSEIT), Vol.2, No.1, February 2012
70
Proposed algorithm to retrieve the related face images:
1. Repeat the same procedure as explained above for the query image.
2. Apply different distance measures to retrieve the relevant faces from the database.
3. Display the relevant faces based on minimum distance criteria.
4. EXPERIMENTAL RESULTS
Recognition performance in terms of average recognition rate and recognition time of the
proposed face recognition system is tested by conducting experiments on yale data base. A face
database [6]
test set was constructed by selecting 100 images of 10 individuals, ten images per
person. These images of a person used for training and testing. the experimental results are
tabulated in Table 1. Since the recognition accuracy of the sub-pattern image, several sizes of
sub-pattern images are used in our experiments as shown below: 56×46(S=4), 28×23(S=16),
14×23(S=32), 7×23(S=64), and 4×23(S=112). Results have been presented in two models. First
model is named as hybrid approach1. In this, vertical, horizontal, and whitened are considered.
Second model is named as hybrid approach2. In this, eigen values, vertical, horizontal, and
whitened are considered.
4.1. Feature selection
Feature selection procedure is shown in Figure 3.
03 JAN 2012 QIS College of Engineering &
Technology ,ONGOLE
24
Selecta querySelecta querySelecta querySelecta querySelecta querySelecta querySelecta querySelecta query
Figure 3. Query Image
From the query image feature is taken based on sub-pattern method. In this paper we take only
64 feature of this query image. That may be depends up on the sub-parts of this image(S=16). For
each sub-pattern we consider four positive eigenvectors that is largest eigenvector of the sub-part.
Comparative performance of all training global feature with this query image finally recognized
results images with top left image as query image.](https://image.slidesharecdn.com/2112ijcseit07-180824090734/85/FACE-RECOGNITION-USING-DIFFERENT-LOCAL-FEATURES-WITH-DIFFERENT-DISTANCE-TECHNIQUES-4-320.jpg)
![International Journal of Computer Science, Engineering and Information Technology (IJCSEIT), Vol.2, No.1, February 2012
73
1 2 3 4 5 6 7 8 9 10
30
40
50
60
70
80
90
100
Number of Top recognized images
RecognitionRate(%)
Diagonal+Eigenvalues+Vertical+Horizontal
Manhattan Distance
Weighted angle Distance
Minkowski Distance
Figure 6. Comparative recognition rates for Hybrid approach2
Results are indicating that Hybrid approach2 give better result with Minkowski distance
compared to hybrid approach1 with all distance measures.
2.10. Recognition Time
Face recognition using different local features with different distance measures is presented. Time
taken for generating the features for the entire database is 52.98 seconds for hybrid approach 1.
Recognition time for Minkowski, Weighted angle and Manhattan are 0.46, 0.45, 0.43seconds
respectively. Similarly database time for hybrid approach2 is 51.84 seconds. Recognized time for
Minkowski, Weighted angle and Manhattan are 0.45, 0.42, 0.38 seconds respectively.
3. CONCLUSIONS
Face recognition using different local features with different distance measures are presented in
this paper. Global feature vectors are generated based on eigenvalues, eigenvectors and diagonal
values with whitened features. Local features are based on vertical and horizontal sub pattern
techniques. Proposed method with Minkowski distance gives better results in term of average
recognized rate and retrieval time compared to the existing methods.
ACKNOWLEDGEMENTS
Authors would like to thank the Dr.S.Srinivas Kumar, Director R&D Cell, JNTUK, Kakinada for
his valuable thoughts. Authors thank to the Dr.Ch.Srinivasa Rao, Sai Aditya Engineering College,
surampalem for his valuable discussions related to this work.
REFERENCES
[1] W.Arnold. M. Smeulders, M. Worring, S. Satini A. Gupta, R. Jain. Content – Based Image
Retrieval at the end of the Early Years, IEEE Transactions on Pattern analysis and Machine
Intelligence, Vol. 22, No:12, pp 1349-1380 , 2000.
[2] Gupta. Visual Information Retrieval Technology: A Virage erspective, Virage Image Engine. API
Specification, 1997.
[3] Ping-Cheng, Pi-Cheng Tung, A Novel Hybrid Approach Based On Subpattern Technique and
Whitened PCA for Face Recognition, Pattern Recognition 42 (2009), 978-984.](https://image.slidesharecdn.com/2112ijcseit07-180824090734/85/FACE-RECOGNITION-USING-DIFFERENT-LOCAL-FEATURES-WITH-DIFFERENT-DISTANCE-TECHNIQUES-7-320.jpg)
![International Journal of Computer Science, Engineering and Information Technology (IJCSEIT), Vol.2, No.1, February 2012
74
[4] Vytautas Perlibakas Distance measures for PCA based face recognition, Pattern Recognition letters 25
(2004), 711-724
[5] S.I. Choi, Kim, C.H. Choi, Shadow compensation in 2D images for face recognition, Pattern
Recognition 40 (2007) 2118–2125.
[6] M. Turk, Pentland, Eigenfaces for recognition, J. Cognitive Neurosci. 3 (1) (1991) 71–86.
[7] Yale database: http://cvc.yale.edu/projects/yalefaces/yalefaces.html.
[8] J. Yang, D. Zhang, J.Y. Yang, Is ICA significantly better than PCA for face recognition? in:
Proceedings of IEEE International Conference on Computer Vision, vol. 1, 2005, pp. 198–203.
[9] A.J. Bell, T.J. Sejnowski, The independent components of natural scenes are edge filters, Vision Res.
37 (23) (1997) 3327–3338.
Authors
M.Koteswara Rao is currently working as associate professor in ECE Department,
QIS College of Engineering & Technology, Ongole, A.P, India. He received his
M.Tech from JNTUK, Kakinada. He has six years experience in the teaching
undergraduate and post graduate students. His research interests in the area of
content based image retrieval.
K.Veera Swamy is currently Professor in ECE department and Principal of QIS
College of Engineering and Technology, Ongole, A.P, India. He received his Ph.D
from JNTUK, Kakinada. He has fourteen years experience in teaching under
graduate and post graduate students. His research interests are in the areas of image
compression, image watermarking, Face recognition, CBIR, and networking
protocols.
K.Anitha Sheela is currently working as Associate Professor in ECE Department,
JNTUH College of Engineering, Hyderabad, A.P, India. She received her Ph.D
from same Institute. She has done her B.Tech from REC Warangal, & M.Tech from
College of Engineering, Osmania University. She has Thirteen years experience in
teaching under and post graduate students. Her research areas include Signal,
Speech, Image processing and Neural networks.
B.Chandra Mohan is currently working as Professor & HOD in ECE Department,
Bapatla Engineering College, Bapatla, A.P, India. He received his Ph.D from JNTU
College of Engineering, Kakinada, A.P, India; He has eighteen years experience in
teaching under and post graduate students. His research interests are in the areas of
Watermarking and communications.](https://image.slidesharecdn.com/2112ijcseit07-180824090734/85/FACE-RECOGNITION-USING-DIFFERENT-LOCAL-FEATURES-WITH-DIFFERENT-DISTANCE-TECHNIQUES-8-320.jpg)
FACE RECOGNITION USING DIFFERENT LOCAL FEATURES WITH DIFFERENT DISTANCE TECHNIQUES
- 1. International Journal of Computer Science, Engineering and Information Technology (IJCSEIT), Vol.2, No.1, February 2012 DOI : 10.5121/ijcseit.2012.2107 67 FACE RECOGNITION USING DIFFERENT LOCAL FEATURES WITH DIFFERENT DISTANCE TECHNIQUES M.Koteswara Rao1 , K.Veera Swamy2 , K.Anithasheela3 and B.Chandra Mohan4 1 Associate Professor, QIS College of Engineering & Technology, Ongole, A.P, India, [email protected] 2 Professor, QIS College of Engineering & Technology, Ongole, A.P, India, [email protected] 3 Associate Professor, JNTUH, Hyderabad, A.P, India, [email protected] 4 Professor, Bapatla Engineering College, A.P, India, [email protected] ABSTRACT A face recognition system using different local features with different distance measures is proposed in this paper. Proposed method is fast and gives accurate detection. Feature vector is based on Eigen values, Eigen vectors, and diagonal vectors of sub images. Images are partitioned into sub images to detect local features. Sub partitions are rearranged into vertically and horizontally matrices. Eigen values, Eigenvector and diagonal vectors are computed for these matrices. Global feature vector is generated for face recognition. Experiments are performed on benchmark face YALE database. Results indicate that the proposed method gives better recognition performance in terms of average recognized rate and retrieval time compared to the existing methods. KEYWORDS Vertical centered, Horizontal centered, Eigen vector, Eigen values and distance methods. 1. INTRODUCTION Face recognition [1] is a computer application for identifying and verifying a person from the face database. The face recognition system is generally used in security purpose and can be compared to other techniques such as fingerprint, iris and signature. More face recognition methods are identify face extract feature, or from an image of the database. These methods may analyze the relative size, shape, position of the database images. These extracted features [5] are used to search for other relevant images. So far face recognition methods developed can be classified as holistic method or local feature method. First method is appearance based technique, which analyze the distribution of individual faces in face space for holistic features. It can be done by using global and local features[4]. Concentration on dynamic link matching or graph matching is considered in local feature [1] method. In global feature method or holistic method concentration on eigenfaces or similar appearance such as Principal Component Analysis (PCA) is considered. PCA approach is mainly concentrated on dimensionality reduction. This scheme is based on linearly projecting the image space to a lower dimensionality space that is also known as Eigen space. Second method is feature based technique [5], which is concentrated on dimensionality of input image as well as images in face database. In face recognition system dimensionality reduction [7] is an essential technique. Block diagram of face recognition system is shown in Figure 1.
- 2. International Journal of Computer Science, Engineering and Information Technology (IJCSEIT), Vol.2, No.1, February 2012 68 Figure 1. Block diagram of face recognition. 2. DISTANCE MEASURES Several distance measures are useful for face recognition. Manhattan distance, weighted angle based Distance, and Minkowski Distance are considered in this work. 2.1. Manhattan Distance Manhattan distance, also known as the L1-distance, between two points in an Euclidean space fixed Cartesian coordinate system sum of the lengths of the projections of the line segment between the points onto the coordinate axes. Notice that the Manhattan distance depends on the choice on the rotation of the coordinate system, but does not depend on the translation of the coordinate system or its reflection with respect to a coordinate axis. Manhattan distance is also known as city block distance. ∑ = −== = n i iip yxZYLZYd 1 ||),(),( 1 Where n is the number of variables, and Xi and Yi are the values of the ith variable, at points Y and Z respectively. 2.2. Weighted angle Distance Cosine similarity [4] is a measure of similarity between two vectors by measuring the cosine of the angle between them. The cosine of 0 is 1, and less than 1 for any other angle; the lowest value of the cosine is -1. The cosine of the angle between two vectors thus determines whether two vectors are pointing in roughly the same direction. Cosine of two vector s can be written as Similarity = ∑ ∑∑ ∑ == = == N I I N I I N I II II YQ YQ YQ YQ YQ 1 2 1 2 1 )(*)( * * ||||.|||| . )cos(θ The resulting similarity ranges from −1 meaning exactly opposite, to 1 meaning exactly the same, with 0 usually indicating independence, and in-between values indicating intermediate similarity or dissimilarity. For text matching, the attribute vectors A and B are usually the term frequency
- 3. International Journal of Computer Science, Engineering and Information Technology (IJCSEIT), Vol.2, No.1, February 2012 69 vectors of the documents. The cosine similarity can be seen as a method of normalizing document length during comparison. 2.3. Minkowski Distance Minkowski distance is concentrated on Euclidean [4] space, which can be considered as a generalization of both Euclidean and Manhattan distance for getting more recognition efficiency. Minkowski distance is based on factor p. ),.....,( 21 nxxxp = Minkowski diastance is typically used with p being 1 or 2. In the limiting case of p reaching infinity we obtain the chebyshev distance. ∑= −== n i pip iip yxZYLZYd 1 / )||(),(),( Minkowski distance is often used when variables are measured on ration scales with absolute zero value. Variables with a wider range can overpower the result. Figure 2 Minkowski distance 3. PROPOSED ALGORITHM Proposed algorithm to get feature vector for the database is given below: 1. Each image is partitioned into sub images. D= r1*r2/k. r1&r2 are the image rows & columns. K is the equally sized. 2. Convert the each sub image into column data matrix. Each of them can be expressed in the order of a D-by-N. Ci = {ci1+ci2+ci3+…ciN} with i = 1, 2,…….. K. here N is the total number of images 3. Calculate mean value for each sub image. 4. Subtract the mean value from column data matrix of each sub image then obtain vertically centered column data matrix Cvi = {ĉi1+ĉi2+ĉi3+…ĉiN} with i = 1, 2,……...K. 5. Rearrange the elements to get square matrix. 6. Collect Eigen values, Eigen vectors, and diagonal values of the square matrix PVi = {Pi1+ Pi2 +Pi3+……+ PiL} with i = 1, 2….S. here L is the feature of the sub image. Then obtain training data base matrix Gvi = PVi T Cvi = {Gi1+ Gi2+……+ GiN} = 1, 2…K. 7. Repeat the same procedure row data matrix. 8. Reduce the feature size as per the requirement. GVj = (G1j T ,G2J T ,……GSj T )T , j = 1,2,..N. 9. Convert the feature into global feature by converting into a single value.
- 4. International Journal of Computer Science, Engineering and Information Technology (IJCSEIT), Vol.2, No.1, February 2012 70 Proposed algorithm to retrieve the related face images: 1. Repeat the same procedure as explained above for the query image. 2. Apply different distance measures to retrieve the relevant faces from the database. 3. Display the relevant faces based on minimum distance criteria. 4. EXPERIMENTAL RESULTS Recognition performance in terms of average recognition rate and recognition time of the proposed face recognition system is tested by conducting experiments on yale data base. A face database [6] test set was constructed by selecting 100 images of 10 individuals, ten images per person. These images of a person used for training and testing. the experimental results are tabulated in Table 1. Since the recognition accuracy of the sub-pattern image, several sizes of sub-pattern images are used in our experiments as shown below: 56×46(S=4), 28×23(S=16), 14×23(S=32), 7×23(S=64), and 4×23(S=112). Results have been presented in two models. First model is named as hybrid approach1. In this, vertical, horizontal, and whitened are considered. Second model is named as hybrid approach2. In this, eigen values, vertical, horizontal, and whitened are considered. 4.1. Feature selection Feature selection procedure is shown in Figure 3. 03 JAN 2012 QIS College of Engineering & Technology ,ONGOLE 24 Selecta querySelecta querySelecta querySelecta querySelecta querySelecta querySelecta querySelecta query Figure 3. Query Image From the query image feature is taken based on sub-pattern method. In this paper we take only 64 feature of this query image. That may be depends up on the sub-parts of this image(S=16). For each sub-pattern we consider four positive eigenvectors that is largest eigenvector of the sub-part. Comparative performance of all training global feature with this query image finally recognized results images with top left image as query image.
- 5. International Journal of Computer Science, Engineering and Information Technology (IJCSEIT), Vol.2, No.1, February 2012 71 03 JAN 2012 QIS College of Engineering & Technology ,ONGOLE 25 Recognized resultRecognized resultRecognized resultRecognized resultRecognized resultRecognized resultRecognized resultRecognized result Figure 4. Recognized result 4.2. Recognized rate Comparative performance in terms of average recognized rate is shown in Figure 5 and Figure 6 respectively 4.2.1 Hybrid approach1 experimental results Table 1. Recognized efficiency on face database for hybrid approach1. Feature Technique Distance Measure 1 3 5 7 10 Hybrid approach1 (Vertical+Horizontal+ Whitened) Manhattan Distance 100 50 41 37.85 35.75 Weighted angle Distance 100 99.16 96.5 87.85 79.25 Minkowski Distance 100 99.16 95 88.57 80.75
- 6. International Journal of Computer Science, Engineering and Information Technology (IJCSEIT), Vol.2, No.1, February 2012 72 1 2 3 4 5 6 7 8 9 10 30 40 50 60 70 80 90 100 Number of Top recognized images Recognitionrate(%) DWT+Vertical+Horizontal+Whitened Manhattan Distance Weighted angle Distance Minkowski Distance Figure 5. Comparative recognition rates for Hybrid approach1 4.2.2. Hybrid approach2 experimental results Table 2. Recognized efficiency on face database for hybrid approach 2. Feature Technique Distance Measure 1 3 5 7 10 Hybrid approach2 (Eigenvalue + Vertical+ Horizontal+Whitened) (PROPOSED) Manhattan Distance 100 53.3 43.50 36.78 33.25 Weighted angle Distance 100 95.83 88 86.78 74.25 Minkowski Distance 100 100 96 90 82
- 7. International Journal of Computer Science, Engineering and Information Technology (IJCSEIT), Vol.2, No.1, February 2012 73 1 2 3 4 5 6 7 8 9 10 30 40 50 60 70 80 90 100 Number of Top recognized images RecognitionRate(%) Diagonal+Eigenvalues+Vertical+Horizontal Manhattan Distance Weighted angle Distance Minkowski Distance Figure 6. Comparative recognition rates for Hybrid approach2 Results are indicating that Hybrid approach2 give better result with Minkowski distance compared to hybrid approach1 with all distance measures. 2.10. Recognition Time Face recognition using different local features with different distance measures is presented. Time taken for generating the features for the entire database is 52.98 seconds for hybrid approach 1. Recognition time for Minkowski, Weighted angle and Manhattan are 0.46, 0.45, 0.43seconds respectively. Similarly database time for hybrid approach2 is 51.84 seconds. Recognized time for Minkowski, Weighted angle and Manhattan are 0.45, 0.42, 0.38 seconds respectively. 3. CONCLUSIONS Face recognition using different local features with different distance measures are presented in this paper. Global feature vectors are generated based on eigenvalues, eigenvectors and diagonal values with whitened features. Local features are based on vertical and horizontal sub pattern techniques. Proposed method with Minkowski distance gives better results in term of average recognized rate and retrieval time compared to the existing methods. ACKNOWLEDGEMENTS Authors would like to thank the Dr.S.Srinivas Kumar, Director R&D Cell, JNTUK, Kakinada for his valuable thoughts. Authors thank to the Dr.Ch.Srinivasa Rao, Sai Aditya Engineering College, surampalem for his valuable discussions related to this work. REFERENCES [1] W.Arnold. M. Smeulders, M. Worring, S. Satini A. Gupta, R. Jain. Content – Based Image Retrieval at the end of the Early Years, IEEE Transactions on Pattern analysis and Machine Intelligence, Vol. 22, No:12, pp 1349-1380 , 2000. [2] Gupta. Visual Information Retrieval Technology: A Virage erspective, Virage Image Engine. API Specification, 1997. [3] Ping-Cheng, Pi-Cheng Tung, A Novel Hybrid Approach Based On Subpattern Technique and Whitened PCA for Face Recognition, Pattern Recognition 42 (2009), 978-984.
- 8. International Journal of Computer Science, Engineering and Information Technology (IJCSEIT), Vol.2, No.1, February 2012 74 [4] Vytautas Perlibakas Distance measures for PCA based face recognition, Pattern Recognition letters 25 (2004), 711-724 [5] S.I. Choi, Kim, C.H. Choi, Shadow compensation in 2D images for face recognition, Pattern Recognition 40 (2007) 2118–2125. [6] M. Turk, Pentland, Eigenfaces for recognition, J. Cognitive Neurosci. 3 (1) (1991) 71–86. [7] Yale database: http://cvc.yale.edu/projects/yalefaces/yalefaces.html. [8] J. Yang, D. Zhang, J.Y. Yang, Is ICA significantly better than PCA for face recognition? in: Proceedings of IEEE International Conference on Computer Vision, vol. 1, 2005, pp. 198–203. [9] A.J. Bell, T.J. Sejnowski, The independent components of natural scenes are edge filters, Vision Res. 37 (23) (1997) 3327–3338. Authors M.Koteswara Rao is currently working as associate professor in ECE Department, QIS College of Engineering & Technology, Ongole, A.P, India. He received his M.Tech from JNTUK, Kakinada. He has six years experience in the teaching undergraduate and post graduate students. His research interests in the area of content based image retrieval. K.Veera Swamy is currently Professor in ECE department and Principal of QIS College of Engineering and Technology, Ongole, A.P, India. He received his Ph.D from JNTUK, Kakinada. He has fourteen years experience in teaching under graduate and post graduate students. His research interests are in the areas of image compression, image watermarking, Face recognition, CBIR, and networking protocols. K.Anitha Sheela is currently working as Associate Professor in ECE Department, JNTUH College of Engineering, Hyderabad, A.P, India. She received her Ph.D from same Institute. She has done her B.Tech from REC Warangal, & M.Tech from College of Engineering, Osmania University. She has Thirteen years experience in teaching under and post graduate students. Her research areas include Signal, Speech, Image processing and Neural networks. B.Chandra Mohan is currently working as Professor & HOD in ECE Department, Bapatla Engineering College, Bapatla, A.P, India. He received his Ph.D from JNTU College of Engineering, Kakinada, A.P, India; He has eighteen years experience in teaching under and post graduate students. His research interests are in the areas of Watermarking and communications.