C. Sanderson and K.K. Paliwal, "Polynomial Features for Robust Face Authentication", Proc. ICIP, Rochester, 2002, pp. 997-1000 (Vol. 3).  Home/Search   Document Details and Download   Summary   Related Articles   Check

No context found.

C. Sanderson and K.K. Paliwal, "Polynomial Features for Robust Face Authentication", Proc. ICIP, Rochester, 2002, pp. 997-1000 (Vol. 3).

No context found.

Sanderson, C. and Paliwal, K. K.: Polynomial Features for Robust Face Authentication. Proc. International Conf. on Image Processing, Rochester, New York, 2002, pp. 997-1000 (Vol. 3). 912, 913

No context found.

C. Sanderson and K. K. Paliwal, "Polynomial Features for Robust Face Authentication", Proc. International Conf. Image Processing, Rochester, New York, 2002.

No context found.

Sanderson, C., and Paliwal, K.K.: Polynomial Features for Robust Face Authentication. Proc. International Conf. on Image Processing, Rochester, New York, 2002, pp. 997-1000 (Vol. 3).

No context found.

Sanderson, C. and Paliwal, K. K.: Polynomial Features for Robust Face Authentication. Proc. International Conf. on Image Processing, Rochester, New York, 2002, pp. 997-1000 (Vol. 3). 912, 913

No context found.

C. Sanderson and K. K. Paliwal, "Polynomial Features for Robust Face Authentication", Proc. International Conf. Image Processing, Rochester, New York, 2002.

No context found.

C. Sanderson and K.K. Paliwal, "Polynomial Features for Robust Face Authentication", Proc. ICIP, Rochester, 2002, pp. 997-1000 (Vol. 3).

No context found.

C. Sanderson and K.K. Paliwal, "Polynomial Features for Robust Face Authentication", Proceedings of International Conference on Image Processing, Rochester, New York, 2002, pp. 997-1000 (Vol. 3).

No context found.

C. Sanderson and K.K. Paliwal, "Polynomial Features for Robust Face Authentication", Proceedings of International Conference on Image Processing, Rochester, New York, 2002, pp. 997-1000 (Vol. 3).

....fadg0, faks0, fcft0, fcmh0, mstk0, mtas1, mtmr0 and mwbt0 (i.e. 4 female and 4 male) are to be used only for impostor tests; this leaves 35 subjects for true claimant tests. In total, there are 1120 (35 8) impostor and 140 (35 4) true claimant tests. Publications using this protocol: [12, 13, 14]. 5.3 Proposed Protocol II: A Priori Performance Type A In this protocol, Session 1 of the database is used for training the client models, Session 2 is used to find the a priori decision threshold (or parameters for an equivalent decision mechanism) and Session 3 is used to find the final ....

....or organization outside of IDIAP. However, it is permitted to publish upto 10 video frames (whether original or processed) of any person or of all the persons in any publication. 2. Any publication resulting from the use of the VidTIMIT database (whether in whole or in part) must reference [12]. 3. The VidTIMIT database (whether in whole or in part) cannot be incorporated into any other database. 4. The VidTIMIT database is provided as is. There is no warranty (whether expressed or implied) regarding fitness for any purpose. 7 Acknowledgments My thanks go to all the volunteers for ....

C. Sanderson and K. K. Paliwal, "Polynomial Features for Robust Face Authentication", Proc. International Conf. Image Processing, Rochester, New York, 2002.

....contains D eigenvectors (with largest corresponding eigenvalues) of the training data covariance matrix, and f is the mean of training face vectors. PCA derived features have been shown to be sensitive to changes in the illumination direction causing rapid degradation in verification performance [29]. In the proposed enhanced PCA approach , a given face image is processed using recently proposed DCT mod2 feature extraction [29] to produce pseudo image F , which is then used in place of F by traditional PCA feature extraction. Since DCT mod2 feature vectors are robust to illumination ....

....face vectors. PCA derived features have been shown to be sensitive to changes in the illumination direction causing rapid degradation in verification performance [29] In the proposed enhanced PCA approach , a given face image is processed using recently proposed DCT mod2 feature extraction [29] to produce pseudo image F , which is then used in place of F by traditional PCA feature extraction. Since DCT mod2 feature vectors are robust to illumination changes, features obtained via the enhanced PCA should also be robust to illumination changes. Formally, the pseudo image is constructed ....

[Article contains additional citation context not shown here]

C. Sanderson and K.K. Paliwal, "Polynomial Features for Robust Face Authentication", Proc. ICIP, Rochester, 2002, pp. 997-1000 (Vol. 3).

....then propose several new feature extraction methods which build from the 2D DCT. The performance of the traditional and proposed feature extraction techniques is compared in Section 4, using artificial and real life illumination direction changes. This report is an extended and revised version of [53]. The reader may also be interested in related works: 7, 56] 2 Summary of Past Face Recognition Approaches This section presents a concise review of previous approaches to automatic face recognition. It goes into detail with the most important and or popular approaches; the reader is also ....

C. Sanderson and K. K. Paliwal, "Polynomial Features for Robust Face Authentication", Proc. International Conf. Image Processing, Rochester, 2002. pp. 997-1000 (Vol. 3).

....it is often assumed that the detection step has been performed perfectly, however, this is not realistic. In this study, results are given for perfect detection as well as in more realistic conditions, i.e. using an automatic face detector. 3 Feature Extraction In DCT mod2 feature extraction [14] a given face image is analyzed on a block by block basis; each block is N N (here we use N = 8) and overlaps neighboring blocks by 50 . Each block is decomposed in terms of 2D Discrete Cosine Transform (DCT) basis functions [7] A feature vector for each block is then constructed as: x = ....

....DCT feature extraction [4] the first three DCT coefficients are replaced by their We use two image sizes: 40 32 and 80 64 (rows columns) respective horizontal and vertical deltas in order to reduce the effects of illumination direction changes. In this study we use M=15 (choice based on [14]) resulting in an 18 dimensional feature vector for each block. Since DCT mod2 feature extraction for a given block is only possible when the block has vertical and horizontal neighbours, processing an image which has Y rows and X columns results in (2 Y 3) 2 3) feature vectors; thus for ....

Sanderson, C., and Paliwal, K.K.: Polynomial Features for Robust Face Authentication. Proc. International Conf. on Image Processing, Rochester, New York, 2002, pp. 997-1000 (Vol. 3).

....set is more robust to compression artefacts and white Gaussian noise. To keep consistency with traditional matrix notation, pixel locations (and image sizes) throughout this chapter are described using the row(s) first, followed by the column(s) Publications resulting from this research: [131, 132, 133, 134, 136, 137, 138]. In UBM alt normalization, both the client and the impostor models are generated using the EM algorithm, which is in contrast to UBM normalization where the client models are generated by adapting the impostor model. 5.2 Summary of Past Face Recognition Approaches This section presents a ....

C. Sanderson and K. K. Paliwal, "Polynomial Features for Robust Face Authentication", Proc. International Conf. Image Processing, Rochester, 2002. pp. 997-1000 (Vol. 3).

....face authentication systems in a common framework i.e. classifier, database, controlled image corruption via an illumination change and compression artefacts. The four systems differ in the feature extraction method used: eigenfaces (PCA) 8] 2 D DCT [9] 2 D Gabor wavelets [10] and DCT mod2 [11]. The rest of the paper is organized as follows. In Section 2 we briefly review the feature extraction methods. In Section 3, we describe the Gaussian Mixture Model (GMM) based classifier which shall be used as the basis for experiments. In Section 4 we describe the two normalization approaches ....

....used to form an # dimensional feature vector (typically, # ###) The DCT mod2 approach is similar to 2 D DCT. The main difference is that the feature vector for each block also contains polynomial coefficients based on a subset of 2 D DCT coefficients extracted from spatially neighbouring blocks [11]. The dimensionality of a DCT mod2 feature vector is # ##. In 2 D Gabor wavelet feature extraction, a coarse rectangular grid is placed over a given image. At each node of the grid, the image is analyzed by a set of biologically inspired 2 D Gabor wavelets [10] differing in orientation and ....

[Article contains additional citation context not shown here]

C. Sanderson and K. K. Paliwal, "Polynomial Features for Robust Face Authentication", Proc. Int. Conf. Image Processing, Rochester, New York, 2002.

....fadg0, faks0, fcft0, fcmh0, mstk0, mtas1, mtmr0 and mwbt0 (i.e. 4 female and 4 male) are to be used only for impostor tests; this leaves 35 subjects for true claimant tests. In total, there are 1120 (35 4 8) impostor and 140 (35 4) true claimant tests. Publications using this protocol: [12, 13, 14]. 5.3 Proposed Protocol II: A Priori Performance Type A In this protocol, Session 1 of the database is used for training the client models, Session 2 is used to nd the a priori decision threshold (or parameters for an equivalent decision mechanism) and Session 3 is used to nd the nal ....

....or organization outside of IDIAP. However, it is permitted to publish upto 10 video frames (whether original or processed) of any person or of all the persons in any publication. 2. Any publication resulting from the use of the VidTIMIT database (whether in whole or in part) must reference [12]. 3. The VidTIMIT database (whether in whole or in part) cannot be incorporated into any other database. 4. The VidTIMIT database is provided as is. There is no warranty (whether expressed or implied) regarding tness for any purpose. 7 Acknowledgments My thanks go to all the volunteers for ....

C. Sanderson and K. K. Paliwal, \Polynomial Features for Robust Face Authentication", Proc. International Conf. Image Processing, Rochester, New York, 2002.

....the variation of the contribution of each block. Results show that depending on the window more robust to compression artefacts and white Gaussian noise. throughout this chapter are described using the row(s) first, followed by the column(s) Publications resulting from this research: 127] [129], 130] 131] 132] 133] and [134] In UBM alt normalization, both the client and the impostor models are generated using the EM algorithm, which is in contrast to UBM normalization where the client models are generated by adapting the impostor model. 5.2 Summary of Past Face Recognition ....

C. Sanderson and K. K. Paliwal, "Polynomial Features for Robust Face Authentication", Proc. International Conf. Image Processing, Rochester, New York, 2002.

No context found.

C. Sanderson and K. Paliwal, "Polynomial Features for Robust Face Authentication, " Proceedings of International Conference on Image Processing, vol. 3, pp. 997-- 1000, 2002.

No context found.

C. Sanderson and K. Paliwal, \Polynomial features for robust face authentication," in Proceedings of International Conference on Image Processing, vol. 3, 2002, pp. 997-1000.

No context found.

C. Sanderson and K. Paliwal, "Polynomial Features for Robust Face Authentication, " Proceedings of International Conference on Image Processing, vol. 3, pp. 997-- 1000, 2002.

Online articles have much greater impact   More about CiteSeer.IST   Add search form to your site   Submit documents   Feedback  

CiteSeer.IST - Copyright Penn State and NEC