Citation Context

... was an LDA algorithm developed by Zhao and Chellapa [17]. Of the top performing algorithms in FERET, this is the one based upon the oldest and best understood subspace projection technique after PCA =-=[4, 1]-=-. For both these reasons, a similar LDA algorithm has been chosen for our study. Stepping back from face recognition, characterizing the performance of computer vision algorithms has been an ongoing c...

Citation Context

... associated with changes in the test data. As a baseline algorithm, FERET used an Eigenface algorithm that represented the line of classification algorithms based upon PCA that arose from the work by =-=[11, 10]-=-. The PCA algorithm used here is for all intents and purposes equivalent to the Eigenface algorithm used in FERET. One of the top performing algorithms in the FERET evaluation was an LDA algorithm dev...

Citation Context

...e make the source code for the algorithms, scoring procedures and Monte Carlo study available in the hopes others will extend this comparison to newer algorithms. 1. Introduction The FERET evaluation =-=[12]-=- established a common data set and a common testing protocol for evaluating semiautomated and automated face recognition algorithms. It illustrated how much can be accomplished in a well coordinated c...

Citation Context

... associated with changes in the test data. As a baseline algorithm, FERET used an Eigenface algorithm that represented the line of classification algorithms based upon PCA that arose from the work by =-=[11, 10]-=-. The PCA algorithm used here is for all intents and purposes equivalent to the Eigenface algorithm used in FERET. One of the top performing algorithms in the FERET evaluation was an LDA algorithm dev...

Citation Context

...tem and probe images are novel images to be recognized. The testing data used in this study consists of � images for each of � distinct individuals. Initially, we endeavored to design a bootstrapp=-=ing [5]-=- study, but difficulties described below led us to instead favor permuting probe and gallery choices. Permuting the images selected to serve as gallery and probe generates sample gallery and probe set...

Citation Context

...e is for all intents and purposes equivalent to the Eigenface algorithm used in FERET. One of the top performing algorithms in the FERET evaluation was an LDA algorithm developed by Zhao and Chellapa =-=[17]-=-. Of the top performing algorithms in FERET, this is the one based upon the oldest and best understood subspace projection technique after PCA [4, 1]. For both these reasons, a similar LDA algorithm h...

Citation Context

...ext books often state that Ï is found by solving the general eigenvector problem [4]: Ì (9) Å � Ï �£Å ÏÏ (11) This is true, but provides no insight into why. Nor is it always the best way =-=solve for Ï [18]. We have written -=-a report [8] illustrating the underlying geometry at work and filling out the solution method used in [18]. Projecting an image Ý into LDA subspace yields Ý : Ý � ÏÝ � Ï� � ÝsÜ � (12...

Citation Context

...oth these reasons, a similar LDA algorithm has been chosen for our study. Stepping back from face recognition, characterizing the performance of computer vision algorithms has been an ongoing concern =-=[7, 9]-=- and more is certainly being done in this area each year. In comparison, however, far more is written each year about new and different algorithms. See [14, 15] for recent surveys of face recognition ...

Citation Context

... such sampling done by brute force retraining on each sample, the computational burden would be staggering. In the past we have studied variation in both PCA and LDA performance subject to retraining =-=[16]-=-. In future we will investigate ways to adapt our methodology efficiently to questions involving changes to the training data. Since it is desirable to have no overlap between training and test data, ...

Citation Context

...n written about using modern statistical methods [2] to measure uncertainty in performance measures. 1 http://www.cs.colostate.edu/evalfacerec/ One notable exception is the work by Micheals and Boult =-=[13]-=-. Micheals and Boult use a statistical technique to derive mean and standard deviation estimates for recognition rates at different ranks. They compare a standard PCA algorithm to two algorithms from ...

Citation Context

...on algorithms has been an ongoing concern [7, 9] and more is certainly being done in this area each year. In comparison, however, far more is written each year about new and different algorithms. See =-=[14, 15]-=- for recent surveys of face recognition algorithms. Thus, while the literature on algorithms is vast, little has been written about using modern statistical methods [2] to measure uncertainty in perfo...

Citation Context

... found by solving the general eigenvector problem [4]: Ì (9) Å � Ï �£Å ÏÏ (11) This is true, but provides no insight into why. Nor is it always the best way solve for Ï [18]. We have writt=-=en a report [8] illustrating the -=-underlying geometry at work and filling out the solution method used in [18]. Projecting an image Ý into LDA subspace yields Ý : Ý � ÏÝ � Ï� � ÝsÜ � (12) Training images must be part...

Citation Context

... different algorithms. See [14, 15] for recent surveys of face recognition algorithms. Thus, while the literature on algorithms is vast, little has been written about using modern statistical methods =-=[2]-=- to measure uncertainty in performance measures. 1 http://www.cs.colostate.edu/evalfacerec/ One notable exception is the work by Micheals and Boult [13]. Micheals and Boult use a statistical technique...

Citation Context

...on algorithms has been an ongoing concern [7, 9] and more is certainly being done in this area each year. In comparison, however, far more is written each year about new and different algorithms. See =-=[14, 15]-=- for recent surveys of face recognition algorithms. Thus, while the literature on algorithms is vast, little has been written about using modern statistical methods [2] to measure uncertainty in perfo...