際際滷

際際滷Share a Scribd company logo

DT-CWT SUBBAND PARTITIONING FOR FAC RECOGNITION

Download as PPT, PDF
P
Punnam Chandar

This paper presents that prior partitioning of the face image will improve the face recognition performance of Principal Component Analysis.

1 of 20
DT-CWT SUBBAND PARTITIONING FOR FAC RECOGNITION K. Punnam Chandar1 , T. Satya Savithri2 1 Kakatiya University Dept. of E.C.E Warangal, INDIA. 息 ICIIS 2014 2 JNTU Dept. of E.C.E Hyderabad, INDIA.
Outline Motivation The Basic Problem that we studied Previous Work Multi Scale Partitioning Real Wavelets Complex Wavelets Our Contribution & Results OneS Representation Results Conclusion 03/22/15 息 ICIIS 2014 2
Motivation The Basic Problem that we studied Principal Component Analysis [Eigen Faces] PCA is a Statistical Method. PCA extracts the featues from the low frequency content of the face image. Perform Face recongnition. This statistical Method de-emphasizes the high frequency information, available to improve the recognition performace [1-2]. 03/22/15 息 ICIIS 2014 3
Motivation The Basic Problem that we studied Linear Discriminant Analysis [Fisher Faces] LDA finds a linear mapping M that maximizes the Linear Class seperability in the low-dimensional representation of the data. LDA is Suscepetible to over fitting the training data [3]. Fisher Face approach is more senstitive to pose variation & variation in illumination. 03/22/15 息 ICIIS 2014 4
Multi Resolution Wavelet Analysis Previous Work Discrete Wavelet Transform J. T. Chien et al. Proposed Discriminant Wavelet faces. C. J. Chen and J. S. Zhang, Proposed Wavelet Energy as new feature vector. Dual Tree Complex Wavelet Transform: Y. H. Sun and M. H. Du, DT-CWT feature combined with onpp for face recognition. G. Y. Zhang et al., Combination of dual tree complex wavelet transform and SVM for face recognition. 03/22/15 息 ICIIS 2014 5
Multi Resolution Wavelet Analysis Our Previous Work: K. Punnam Chandar et al. Suitability of Complex Wavelets towards face recognition, Highest recognition rate reported (rank-1) 84%. 03/22/15 息 ICIIS 2014 6
Multiscale Partitioning Real Wavelets Real Wavelets Partition the image into four sub bands, Low-Low, Low-High, High-Low, High-High. Inspite of the compact support and efficient computation the DWT suffers from four fundamental, interwined shortcomings, oscillations, shift variance,aliasing and directilnality. 03/22/15 息 ICIIS 2014 7
Scale-one partition using DWT (db2). 03/22/15 息 ICIIS 2014 8
Multiscale Partitioning Complex Wavelets Complex Wavelets on the other hand partition the image into eight sub bands. The short comings of the real wavelets are overcomed. 03/22/15 息 ICIIS 2014 9
Scale-One Partitioning of ORL Database Face Image using DT-CWT. 03/22/15 息 ICIIS 2014 10
Our Contribution Ones Representation The scale-1 partitioning of face image using DT-CWT results in six complex coefficient high frequency sub bands oriented in directions -75, - 45, -15, 15, 45, 75 and two low frequency complex sub bands Low-Low (LL1), Low-Low (LL2). The scale one complex sub band coefficients magnitudes of each sub band are normalized to zero mean and unit Variance. 03/22/15 息 ICIIS 2014 11
Our Contribution Ones Representation The resultant magnitudes of the sub bands coefficients are arranged from low frequency to high frequency as [LL1, LL2, -15, 15, -45, 45, -75, 75]T in to [(8xM/2xN/2), 1] vector, we call this vector as one scale (OneS) Representation. 03/22/15 息 ICIIS 2014 12
Face Recognition Results on ORL Database 03/22/15 息 ICIIS 2014 13 NAME EER VERIFICATION RATE AT 1% FAR 0.1% FAR 0.01 % FAR CWT_ONES 3.58% 92.14% 80.71% 67.86% DB2_ONES 8.95% 74.64% 52.14% 39.29% COIF2_ONE S 7.85% 73.21% 45.36% 28.21% PCA 5.70% 83.44% 67.19% 31.25%
Face Recognition Results on ORL Database 03/22/15 息 ICIIS 2014 14 NAME EER VERIFICATION RATE AT 1% FAR 0.1% FAR 0.01 % FAR CWT_ONES 4.64% 87.14% 69.64% 47.80% DB2_ONES 5.00% 87.14% 67.50% 44.29% COIF2_ONES 4.64% 87.50 70.71% 47.14% PCA 5.70% 83.44% 67.19% 31.25%
DET Curves Comparison of DT-CWT, DWT & PCA 03/22/15 息 ICIIS 2014 15
DET Curves comparison of DT-CWT, DWT & PCA (Normalized Data). 03/22/15 息 ICIIS 2014 16
Conclusion: CWT, DWT Partitioning of face images performed. Novel OneS representation is formed using the sub bands. Further PCA analysis is performed on OneS representation and results are compared. DET Curves are used to compare the performance of OneS representation. Relative Face recognition improvement is 3.7% with OneS representation compared to PCA.03/22/15 息 ICIIS 2014 17
References: 1. Cook, Jamie, Vinod Chandran, and Sridha Sridharan. "Multiscale representation for 3-D face recognition." Information Forensics and Security, IEEE Transactions on 2.3 (2007): 529-536. 2. W. zhao, R. Chellapa, P. J. Phillips, and A. Rosenfeld, Face Recognition: A Literature Survey, ACM comput. Surveys, vol.35, no.4, pp. 399-458, 2003. 3. A. Martinez and A. Kak, PCA versus LDA, IEEE Trans. Pattern Anal. Mach. Intell., vol.23,no.2,pp.228-233, Feb 2001. 4. J. T. Chien and C. C. Wu, Discriminant Wavelet faces and nearest feature classifiers for face recognition, IEEE Trans. Pattern Anal. Mach. Intell., vol. 24, no. 12, pp. 1644-1649, Dec. 2002. 5. L. L. Shen and L. Bai, A review on gabor wavelts for face recognition, Pattern Anal. Appl., vol.9, pp. 273 - 292, 2006. 6. N. G. Kingsbury, Shift invariance properties of the dual tree complex wavelet transform, in Proc. ICASSP99, Phoenix, Az, Mar. 16-19, 1999. 03/22/15 息 ICIIS 2014 18
03/22/15 息 ICIIS 2014 19 Any Queries?
03/22/15 息 ICIIS 2014 20 Thank You.

More Related Content

DT-CWT SUBBAND PARTITIONING FOR FAC RECOGNITION

  • 1. DT-CWT SUBBAND PARTITIONING FOR FAC RECOGNITION K. Punnam Chandar1 , T. Satya Savithri2 1 Kakatiya University Dept. of E.C.E Warangal, INDIA. 息 ICIIS 2014 2 JNTU Dept. of E.C.E Hyderabad, INDIA.
  • 2. Outline Motivation The Basic Problem that we studied Previous Work Multi Scale Partitioning Real Wavelets Complex Wavelets Our Contribution & Results OneS Representation Results Conclusion 03/22/15 息 ICIIS 2014 2
  • 3. Motivation The Basic Problem that we studied Principal Component Analysis [Eigen Faces] PCA is a Statistical Method. PCA extracts the featues from the low frequency content of the face image. Perform Face recongnition. This statistical Method de-emphasizes the high frequency information, available to improve the recognition performace [1-2]. 03/22/15 息 ICIIS 2014 3
  • 4. Motivation The Basic Problem that we studied Linear Discriminant Analysis [Fisher Faces] LDA finds a linear mapping M that maximizes the Linear Class seperability in the low-dimensional representation of the data. LDA is Suscepetible to over fitting the training data [3]. Fisher Face approach is more senstitive to pose variation & variation in illumination. 03/22/15 息 ICIIS 2014 4
  • 5. Multi Resolution Wavelet Analysis Previous Work Discrete Wavelet Transform J. T. Chien et al. Proposed Discriminant Wavelet faces. C. J. Chen and J. S. Zhang, Proposed Wavelet Energy as new feature vector. Dual Tree Complex Wavelet Transform: Y. H. Sun and M. H. Du, DT-CWT feature combined with onpp for face recognition. G. Y. Zhang et al., Combination of dual tree complex wavelet transform and SVM for face recognition. 03/22/15 息 ICIIS 2014 5
  • 6. Multi Resolution Wavelet Analysis Our Previous Work: K. Punnam Chandar et al. Suitability of Complex Wavelets towards face recognition, Highest recognition rate reported (rank-1) 84%. 03/22/15 息 ICIIS 2014 6
  • 7. Multiscale Partitioning Real Wavelets Real Wavelets Partition the image into four sub bands, Low-Low, Low-High, High-Low, High-High. Inspite of the compact support and efficient computation the DWT suffers from four fundamental, interwined shortcomings, oscillations, shift variance,aliasing and directilnality. 03/22/15 息 ICIIS 2014 7
  • 8. Scale-one partition using DWT (db2). 03/22/15 息 ICIIS 2014 8
  • 9. Multiscale Partitioning Complex Wavelets Complex Wavelets on the other hand partition the image into eight sub bands. The short comings of the real wavelets are overcomed. 03/22/15 息 ICIIS 2014 9
  • 10. Scale-One Partitioning of ORL Database Face Image using DT-CWT. 03/22/15 息 ICIIS 2014 10
  • 11. Our Contribution Ones Representation The scale-1 partitioning of face image using DT-CWT results in six complex coefficient high frequency sub bands oriented in directions -75, - 45, -15, 15, 45, 75 and two low frequency complex sub bands Low-Low (LL1), Low-Low (LL2). The scale one complex sub band coefficients magnitudes of each sub band are normalized to zero mean and unit Variance. 03/22/15 息 ICIIS 2014 11
  • 12. Our Contribution Ones Representation The resultant magnitudes of the sub bands coefficients are arranged from low frequency to high frequency as [LL1, LL2, -15, 15, -45, 45, -75, 75]T in to [(8xM/2xN/2), 1] vector, we call this vector as one scale (OneS) Representation. 03/22/15 息 ICIIS 2014 12
  • 13. Face Recognition Results on ORL Database 03/22/15 息 ICIIS 2014 13 NAME EER VERIFICATION RATE AT 1% FAR 0.1% FAR 0.01 % FAR CWT_ONES 3.58% 92.14% 80.71% 67.86% DB2_ONES 8.95% 74.64% 52.14% 39.29% COIF2_ONE S 7.85% 73.21% 45.36% 28.21% PCA 5.70% 83.44% 67.19% 31.25%
  • 14. Face Recognition Results on ORL Database 03/22/15 息 ICIIS 2014 14 NAME EER VERIFICATION RATE AT 1% FAR 0.1% FAR 0.01 % FAR CWT_ONES 4.64% 87.14% 69.64% 47.80% DB2_ONES 5.00% 87.14% 67.50% 44.29% COIF2_ONES 4.64% 87.50 70.71% 47.14% PCA 5.70% 83.44% 67.19% 31.25%
  • 15. DET Curves Comparison of DT-CWT, DWT & PCA 03/22/15 息 ICIIS 2014 15
  • 16. DET Curves comparison of DT-CWT, DWT & PCA (Normalized Data). 03/22/15 息 ICIIS 2014 16
  • 17. Conclusion: CWT, DWT Partitioning of face images performed. Novel OneS representation is formed using the sub bands. Further PCA analysis is performed on OneS representation and results are compared. DET Curves are used to compare the performance of OneS representation. Relative Face recognition improvement is 3.7% with OneS representation compared to PCA.03/22/15 息 ICIIS 2014 17
  • 18. References: 1. Cook, Jamie, Vinod Chandran, and Sridha Sridharan. "Multiscale representation for 3-D face recognition." Information Forensics and Security, IEEE Transactions on 2.3 (2007): 529-536. 2. W. zhao, R. Chellapa, P. J. Phillips, and A. Rosenfeld, Face Recognition: A Literature Survey, ACM comput. Surveys, vol.35, no.4, pp. 399-458, 2003. 3. A. Martinez and A. Kak, PCA versus LDA, IEEE Trans. Pattern Anal. Mach. Intell., vol.23,no.2,pp.228-233, Feb 2001. 4. J. T. Chien and C. C. Wu, Discriminant Wavelet faces and nearest feature classifiers for face recognition, IEEE Trans. Pattern Anal. Mach. Intell., vol. 24, no. 12, pp. 1644-1649, Dec. 2002. 5. L. L. Shen and L. Bai, A review on gabor wavelts for face recognition, Pattern Anal. Appl., vol.9, pp. 273 - 292, 2006. 6. N. G. Kingsbury, Shift invariance properties of the dual tree complex wavelet transform, in Proc. ICASSP99, Phoenix, Az, Mar. 16-19, 1999. 03/22/15 息 ICIIS 2014 18
  • 19. 03/22/15 息 ICIIS 2014 19 Any Queries?
  • 20. 03/22/15 息 ICIIS 2014 20 Thank You.