S. A. Rizvi, P. J. Phillips, and H. Moon. The FERET verification testing protocol for face recognition algorithms. In FG'98, pages 48--53, 1998.  Home/Search   Document Details and Download   Summary   Related Articles   Check

....has recently gained significant attention, especially during the past several years. This is evidenced by the emergence of specific face recognition conferences such as AFGR [24] and AVBPA [23] and systematic empirical evaluation of face recognition techniques (FRT) including the FERET [22, 27] and XM2VTS protocols [18] There are at least two reasons for such a trend: the first is the WenYi Zhao was at the Center for Automation Research while this work was performed. wide range of commercial and law enforcement applications and the second is the availability of feasible technologies ....

....surveillance where the image size of the face area is very small. On the other hand, the subspace LDA system [40] works well with both large and small images, e.g. 96 Theta 84 or 24 Theta 21. It also has the best performance for verification tasks according to the most recent FERET test [27]. 2.1 Two problems in face recognition Despite the successes of many systems [19, 40, 21] based on the FERET test, many issues remain to be addressed. Among those issues the following two are prominent for most systems: 1) the illumination problem, 2) the pose problem. The illumination problem ....

S.A. Rizvi, P.J. Phillips, and H. Moon, "The FERET Verification Testing Protocol for Face Recognition Algorithms," in Proc. International Conference on Automatic Face and Gesture Recognition, 1998.

....and vertical directions) are used to crop the face part from the original image and scale it to any desired resolution. This approach provides the groundtruth for our evaluation by allowing us to separate the issues of localisation and verification (compare with the FERET face recognition test [14]) Fully automatic registration: Automatic alignment of the face images is achieved using a robust form of correlation [7] Given an affine transformation a, the error function expressing the intensity difference between a pixel s in the model image Im and its projection in the probe image Ip is ....

S. A. Rizvi, P. J. Phillips, and H. Moon. The FERET verification testing protocol for face recognition algorithms. In FG'98, pages 48--53, 1998.

....lack robustness. For this reason we have performed face registration semiautomatically. This is the only part of the method in which manual intervention is required and it allows us to separate the issues of localisation and verification recognition (compare with the FERET face recognition test [10]) The procedure is based on manually localised eye positions. Four parameters computed from the eye coordinates (rotation, scaling and translation in the horisontal and vertical directions) are used to crop the face part from the original image and scale it to any desired resolution. Photometric ....

S. A. Rizvi, P. J. Phillips, and H. Moon. The FERET verification testing protocol for face recognition algorithms. In FG'98, pages 48--53, 1998.

....Furthermore, the pose estimation experiment shows that an individual representation is superior to a general representation as used in [1] since the extraction of relevant information is maximized. In an ongoing work we are verifying the recognition ability of our GWN on the FERET database [21] and it will be future work to enhance our pose estimation to a gaze detection system. It will also be future work to evaluate further the abstraction abilities of GWN. Acknowledgment The images used are derived from the Yale Face Database. We thank T. B uhlow and N. Kr uger for valuable ....

S. Rizvi, P. Phillips, and H. Moon. The FERET verification testing protocol for face recognition algorithms. In Int. Conf. on Automatic Face- and Gesture-Recognition, pages 48--53, 1998.

....choice of dimensionality of the subspace is based mainly on the characteristics of the eigenvectors instead of the eigenvalues. Combining PCA and LDA, a subspace LDA face recognition system has been tested in a recent FERET test with a database of around 3,800 testing images among other systems [8, 9]. Although the recent FERET test shows some significant improvement of most algorithms over the last FERET test, the results also reveal that the current systems suffer from performance degradation due to illumination change, 3D rotation etc. In this paper, we propose various methods to improve ....

....can affect the performance of the existing subspace LDA system and then propose methods to improve the robustness of the existing system. 2. 1 The effects on system performance Subspace LDA is a pure statistical approach which has proven to be one of the best for face recognition verification [8, 9]. It first projects the input image into a face subspace which is obtained via PCA 2 , and then feeds the projection coefficients into a LDA classifier. We study the effects of illumination change etc. on the performance of the subspace LDA system through matrix vector analysis. Let us denote a ....

Syed A. Rizvi, P.J. Phillips, H. Moon, "The FERET Verification Testing Protocol for Face Recognition Algorithms", Third International Conference on Automatic Face and Gesture Recognition, Japan, April 1998.

....individuals. 2.6 Head orientation and scene complexity Face orientation relative to the camera, presence and actions of other people, and background conditions may influence face analysis. In the face recognition literature, face orientation has received deliberate attention. The FERET data base [17], for instance, includes both frontal and oblique views, and several specialized data bases have been collected to try to develop methods of face recognition that are invariant to moderate change in face orientation [20] In the face expression literature, use of multiple perspectives is rare and ....

S.R. Rizvi, P.J. Phillips, H. Moon, The Feret verification testing protocol for face recognition algorithms. Proceedings of the Third International Conference on Automatic Face and Gesture Recognition, 48-55, 1998.

....the recognition and return the result of the match to the user. When this level of software interface is used, data from a standard test collection can be loaded into the software and the result can be scored using standard procedures similar to those widely used in areas such as face recognition [2, 3]. In some cases the potential system user will not have an API package but will need to evaluate a COTS (Commercial Off The Shelf) biometric system. In this case a small scale test which provides information about the performance of the system for the users specific application area is needed. ....

....requires that the biometric be used to decide if the input biometric identifies a specific user, are you who you say you are. The identification task requires that the biometric be used to find a matching biometric signature, if it exists, in a database of biometrics such as the FERET test data [2, 3] for face or NIST Special Database 24 [1] for fingerprints. This distinction is important in both the system design and testing. In the verification system design, a relatively slow matching method can be used since only one match is needed to make the decision. In the identification task, the ....

S. Rizvi, P. J. Phillips, and H. Moon. The FERET verification testing protocol for face recognition algorithms. Image and Vision Computing Journal, (to appear) 1999.

....Face recognition is a challenging problem related to both statistical pattern recognition and computer vision. The experiment we conducted here is the application of subspace DCA for face recognition and its comparison with subspace DA system which is one of the best face verification systems [12]. The facesubspace was obtained by training 1038 FERET images from total 444 classes with only 2 to 4 training samples per class. We perform simple normalization to all input face images and the normalized image size is chosen to be 48 Theta 42. Finally the near optimal dimension of the subspace ....

Syed A. Rizvi, P.J. Phillips, H. Moon, "The FERET Verification Testing Protocol for Face Recognition Algorithms, " IEEE AFGR'98, Japan, April 1998.

....last row is for results using the soft metric. has about 2200 persons. The robust version [30] of this system has also been tested on an MPEG 7 content set [31] For comprehensive performance comparison of this algorithm and other algorithms [9, 10, 7, 6, 25] on the FERET test, please refer to [14, 15]. For purposes of comparison, we have tested different algorithms on different face databases. The algorithms we tested are PCA , LDA (both weighted and unweighted) and subspace LDA. The databases we have used are the FERET development dataset [40] USC dataset [11] Olivetti dataset , and ....

....the rank of the ordering of the gallery from a match with a probe, while the y axes represent the fraction of the probes correctly identified in terms of top N match. More recently, some preliminary results show that our system s performance on the task of person verification is very competitive [15]. 3.3 Mugshot Dataset One potential application for face identification algorithms is the electronic mugbook. In an electronic mugbook the gallery consists of digital mugshots of known people and the probe is a digital mugshot of an individual to be identified. In this experiment we give ....

S.A. Rizvi, P.J. Phillips, and H. Moon, "The FERET Verification Testing Protocol for Face Recognition Algorithms," in Proc. Third International Conference on Automatic Face and Gesture Recognition, Nara, Japan, 1998.

....made it difficult to assess the status of FRT for real applications, even though many existing systems reported almost perfect performance on small databases. Measuring the performance of FRT in a framework that models real world settings was one of the three primary goals of the FERET program [166, 167, 168, 5]. The other two goals were advancing FRT and collecting a large database of facial images to support algorithm development and evaluation. The database was collected between August 1993 and July 1996, and consists of 14,126 images of 1199 individuals. The FERET database was divided into ....

....(1199 original sets and 365 duplicate sets) a total of 14,126 images. A development set of 503 sets of images were released to researchers; the remaining images were sequestered by the Government for independent evaluation. 29 5.1. 2 Evaluation For details of the three FERET evaluations see [166, 167, 168, 5]. The results of the Sep96 FERET evaluation, the most recent, will be briefly reviewed here. This evaluation was administered in September 1996 and March 1997. Each algorithm was given two sets of images, a target set and a query set; these are different from the gallery and probe sets used to ....

S. Rizvi, P. J. Phillips, and H. Moon, "The FERET Verification Testing Protocol for Face Recognition Algorithms," Image and Vision Computing, to appear.

....images and a mugshot dataset which has about 2200 persons. The robust version [30] of this system has also been tested on an MPEG 7 content set [31] For comprehensive performance comparison of this algorithm and other algorithms [10] 11] 8] 7] 25] on the FERET test, please refer to [15] [16]. For purposes of comparison, we have tested different algorithms on different face databases. The algorithms we tested are PCA 6 , LDA (both weighted and unweighted) and subspace LDA. The databases we have used are the FERET development dataset [40] USC dataset [12] Olivetti dataset 7 , ....

....the rank of the ordering of the gallery from a match with a probe, while the y axes represent the fraction of the probes correctly identified in terms of top N match. More recently, some preliminary results show that our system s performance on the task of person verification is very competitive [16]. C. Mugshot Dataset One potential application for face identification algorithms is the electronic mugbook. In an electronic mugbook the gallery consists of digital mugshots of known people and the probe is a digital mugshot of an individual to be identified. In this experiment we give ....

S.A. Rizvi, P.J. Phillips, and H. Moon, "The FERET Verification Testing Protocol for Face Recognition Algorithms," in Proc. Third International Conference on Automatic Face and Gesture Recognition, Nara, Japan, 1998.

....application, an algorithm is presented with a face and a claimed identity, and the algorithm must accept or reject the claim. In this paper, we describe the FERET database, the Sep96 FERET evaluation protocol, and present identification results. Verification results are presented in Rizvi et al. [8]. To obtain a robust assessment of performance, algorithms are evaluated against different categories of images. The categories are broken out by lighting changes, people wearing glasses, and the time between the acquisition date of the database image and the image presented to the algorithm. By ....

....universe, some probes are not in the gallery. Both models reflect different and important aspects of face recognition algorithms and report different performance statistics. The open universe models verification applications. The FERET scoring procedures for verification is given in Rizvi et al. [8]. The closed universe model allows one to ask how good an algorithm is at identifying a probe image; the question is not always is the top match correct but is the correct answer in the top n matches This lets one know how many images have to be examined to get a desired level of ....

[Article contains additional citation context not shown here]

S. Rizvi, P. J. Phillips, and H. Moon. The feret verification testing protocol for face recognition algorithms. Image and Vision Computing Journal, (to appear) 1999.

No context found.

S. A. Rizvi, P. J. Phillips, and H. Moon. The FERET verification testing protocol for face recognition algorithms. In FG'98, pages 48--53, 1998.

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