F. Samaria & A. Harter, "Parametrisation of a stochastic model for human face identification", Proc. 2nd IEEE workshop on Applications of Computer Vision, 1994.  Home/Search   Document Not in Database   Summary   Related Articles   Check

.... log N) A kd tree [3] with suitable data can reduce average case expected time to O(N log ) 2.4 Algorithm 3: class constrained test train splits In some applications, the training set is constrained to contain a certain number of items from each class. For instance, the ORL face dataset [8] contains 10 images from each of 40 individuals. The face recognition experiments presented in [6, 9] examine accuracy by varying the size of the training set using 1, 3, or 5 images for each of the 40 classes. In general, we define k to be the number of items required from class k to be in the ....

....running Linux 2.2.1. PARI: http: hasse.mathematik.tu muenchen.de ntsw pari view . uniformly without replacement. Experiment 1 corresponds to Section 2.2 (Algorithm 1: 172) Experiment 2 corresponds to Section 2. 3 (Algorithm 2: 172; R = 5) Experiments 3 and 4 used the ORL dataset [8], with images reduced to 16 16 size (400 items, each with 256 numeric values; 10 items in each of 40 classes) Experiment 3 corresponds to Section 2.4 (Algorithm 3: 8k k = 3) Experiment 4 corresponds to Section 2.5 (Algorithm 4: 8k k = 3; R = 3) These experiments reflect the theoretical ....

F. Samaria and A. Harter. Parametrisation of a stochastic model for human face identification. In Proceedings of IEEE Workshop on Applications on Computer Vision, 1994. ORL database is available at: <www.cam-orl.co.uk/facedatabase.html>.

....valid here. 2. 3 Classifier 3: Stratified 1 NN In some applications, the training set is constrained to contain a certain number of items from each class (typically to reflect the proportions of the classes in S) this is termed stratification (Kohavi, 1995) For instance, the ORL face dataset (Samaria Harter, 1994) contains 10 images from each of 40 individuals. The face recognition experiments presented in (Lawrence et al. 1996; Sim et al. 2000) examine accuracy by varying the size of the training set using 1, 3, or 5 images for each of the 40 classes. In general, we define c to be the number of items ....

Samaria, F., & Harter, A. (1994). Parametrisation of a stochastic model for human face identification. Proceedings of IEEE Workshop on Applications on Computer Vision ORL database is available at: <www.camorl. co.uk/facedatabase.html>.

....features such as forehead, eyes, nose, mouth and chin [69] The observation sequence O is generated from an X Theta Y image using an X Theta L sampling window with X Theta M pixels overlap. Each observation vector is a block of L lines. There is an M line overlap between successive observations [70]. 14 Given c face images for each subject of the training set, the goal of the training stage is to optimize the parameters of the HMM to best describe the observations in the sense of maximizing the probability of the observations given the model. Recognition is carried out by matching the ....

F. Samaria and A. Harter, "Parametrisation of stochastic model for human face identification," in Proceedings of the Second IEEE Workshop on Application of Computer Vision, pp. 138--142, 1994.

....termed ARENA, that satisfies the requirements outlined above and also significantly outperforms PCA based methods on two standard face recognition datasets. 2. Image Datasets and Preprocessing Our results use human face images from two standard datasets: Olivetti Oracle Research Lab (ORL) [22] and FERET [17, 19] ORL consists of 400 frontal faces: 10 tightly cropped images of 40 individuals with only minor variations in pose ( 20 ) illumination and facial expression. The faces are consistently positioned in the image frame, and very little background is visible. FERET contains over ....

....1 summarizes these results. The last two rows of the table present results obtained with ARENA, augmented with the synthetic images (p 2 f0; 1g) Both variants of ARENA outperform all of the reported results. Face recognition using Hidden Markov Models (HMM) on the ORL database is reported in [22]. Their best al 4 This was PCA 2 with Euclidian distance, m = 40, n = 5 (without discarding top eigenvectors) and 10 synthetic images when tested on the ORL dataset. Images per person 1 3 5 Eigenface avg per class 61.4 71.1 74.0 Eigenface one per img 61.4 81.8 89.5 PCA CN 65.8 ....

F. Samaria and A. Harter. Parametrisation of a stochastic model for human face identification. In Proceedings of IEEE Workshop on Applications on Computer Vision, 1994. ORL database is available at: <www.camorl. co.uk/facedatabase.html>.

....as forehead, eyes, nose, mouth and chin [38] The observation sequence O is generated from an X Theta Y image using an X Theta L sampling window with X Theta M pixels overlap (Figure 2) Each observation vector is a block of L lines. There is an M line overlap between successive observations [39]. Figure 2: Image sampling technique for one dimensional HMM Given c face images for each subject of the training set, the goal of the training stage is to optimize the parameters i = A; B; to best describe the observations O = fo 1 ; o 2 ; o T g, in the sense of maximizing P (Oj) ....

F. Samaria and A. Harter, "Parametrisation of stochastic model for human face identification, " in Proceedings of the Second IEEE Workshop on Application of Computer Vision, 1994.

....to the observation vector, while large L increases the probability of cutting across the features. W H L P Figure 2: Face image parameterization and blocks extraction However, the system recognition rate is not very sensitive to variations in L, as long as P is large (P L Gamma 1) In [10] the effect of parameters P and L, together with the effect of the number of states used in the HMM has been extensively discussed. In [6] the observation vectors consist of all the pixel values from each of the blocks, and therefore the dimension of the observation vector is L Theta W (L = 10 ....

F. Samaria and A. Harter, "Parametrisation of stochastic model for human face identification," in Proceedings of the Second IEEE Workshop on Application of Computer Vision, 1994.

....lines. There is an M line overlap between successive observations. The overlapping allows the features to be captured in a manner which is independent of vertical position, while a disjoint partitioning of the image could result in the truncation of features occurring across blocks boundaries. In [36], the effect of different sampling parameters has been discussed. With no overlap, if a small height of the sampling window is used, the segmented data do not correspond to significant facial features. However, as the window height increases there is a higher probability of cutting across the ....

F. Samaria and A. Harter, "Parametrisation of stochastic model for human face identification," in Proceedings of the Second IEEE Workshop on Application of Computer Vision, 1994.

....log N) 7 A kd tree [3] with suitable data can reduce average case expected time to O(N log ) 2.4 Algorithm 3: class constrained test train splits In some applications, the training set is constrained to contain a certain number of items from each class. For instance, the ORL face dataset [8] contains 10 images from each of 40 individuals. The face recognition experiments presented in [6, 9] examine accuracy by varying the size of the training set using 1, 3, or 5 images for each of the 40 classes. In general, we define k to be the number of items required from class k to be in the ....

....Linux 2.2.1. 10 PARI: http: hasse.mathematik.tu muenchen.de ntsw pari view . uniformly without replacement. Experiment 1 corresponds to Section 2.2 (Algorithm 1: 172) Experiment 2 corresponds to Section 2. 3 (Algorithm 2: 172; R = 5) Experiments 3 and 4 used the ORL dataset [8], with images reduced to 16 16 size (400 items, each with 256 numeric values; 10 items in each of 40 classes) Experiment 3 corresponds to Section 2.4 (Algorithm 3: 8k k = 3) Experiment 4 corresponds to Section 2.5 (Algorithm 4: 8k k = 3; R = 3) These experiments reflect the theoretical ....

F. Samaria and A. Harter. Parametrisation of a stochastic model for human face identification. In Proceedings of IEEE Workshop on Applications on Computer Vision, 1994. ORL database is available at: <www.cam-orl.co.uk/facedatabase.html>.

.... for example, 2] Extensions to this technique include low dimensional coding to simplify the template representation and improve the performance of the template matching process (see, for example, the eigenfaces of Turk and Pentland [9] stochastic modeling with Hidden Markov Models (HMMs) [8]) and, elastic face transforms to model the deformation of the face under a rotation in depth ( 11] Neural network based image techniques use an input image representation that is the grey level pixel based image or transformed image which is used as an input to one of a variety of neural ....

....to discriminate between the two persons in some poses. Figure 2: Misclassification of a test view for 5 training views of two persons (see text) For comparison we include some benchmark results obtained by other workers on the same, but individually processed, face databases. Samaria et al. [8] used the HMM implementation, Zhang et al. [12] implemented both the eigenface and elastic matching algorithms and, Achermann et al. [1] used a combination of eigenface, HMM and profile classifiers. Their results are summarised in Table 2. Recognition Algorithm ( Reference HMM Eigenface Elastic ....

F. Samaria and A. Harter. Parametrisation of a stochastic model for human face identification. In Proceedings 2nd IEEE Workshop on Applications of Computer Vision, Sarasota, Florida, December 1994.

....and places face recognition in the context of our visitor identification application. Section 10 concludes by presenting promising directions for future research. 2 Image Datasets and Preprocessing Our results use human face images from two standard datasets: Olivetti Oracle Research Lab (ORL) [20] and FERET [15, 17] ORL consists of 400 frontal faces: 10 tightly cropped images of 40 individuals with only minor variations in pose ( 20 ) illumination and facial expression. The faces are consistently positioned in the image frame, and very little background is visible. FERET contains ....

....1 summarizes these results. The last two rows of the table present results obtained with ARENA, augmented with the synthetic images (p 2 f0; 1g) Both variants of ARENA outperform all of the reported results. Face recognition using Hidden Markov Models (HMM) on the ORL database is reported in [20]. Their best algorithm, with n = 5, obtained an accuracy of 88 , putting it between Lawrence s implementations of PCA 1 and PCA 2. This is inferior to any ARENA variant. 8 Computational Complexity and Storage In this section, we examine the computational complexity of PCA 1, PCA 2, and ARENA, ....

F. Samaria and A. Harter. Parametrisation of a stochastic model for human face identification. In Proceedings of IEEE Workshop on Applications on Computer Vision, 1994. ORL database is available at: <www.camorl. co.uk/facedatabase.html>.

.... [2] Extensions to this technique include low dimensional coding to simplify the template representation and improve the performance of the template matching process (see, for example, the eigenfaces of Turk and Pentland [16] or wavelets, stochastic modeling with Hidden Markov Models (HMMs) [14]) and elastic face transforms to model the deformation of the face under a rotation in depth ( 19] Neural network based image techniques use an input image representation that is the grey level pixel based image or transformed image which is used as an input to one of a variety of neural ....

F. Samaria and A. Harter. Parametrisation of a stochastic model for human face identification. In Proceedings 2nd IEEE Workshop on Applications of Computer Vision, Sarasota, Florida, December 1994.

....of frontal face recognition, and in this domain, techniques based on Principal Components Analysis, popularly termed eigenfaces (Turk Pentland 1991; Pentland, Moghaddam, Starner 1994) have demonstrated good performance. Most published results report experiments on standard datasets such as ORL (Samaria Harter 1994) or FERET (Phillips et al. 1997) Our extensive experiments with ARENA and PCAbased methods on both FERET and ORL datasets appear in (Sim et al. 1999) Here, we present only the results of one such test: Table 1 summarizes a direct comparison of ARENA with the best published face recognition ....

Samaria, F., and Harter, A. 1994. Parametrisation of a stochastic model for human face identication.

No context found.

F. Samaria & A. Harter, "Parametrisation of a stochastic model for human face identification", Proc. 2nd IEEE workshop on Applications of Computer Vision, 1994.

No context found.

F. Samaria and A. Harter, "Parametrisation of a stochastic model for human face identification", Proc. 2nd IEEE workshop on Applications of Computer Vision, 1994.

No context found.

F. Samaria & A. Harter, "Parametrisation of a stochastic model for human face identification", Proc. 2nd IEEE workshop on Applications of Computer Vision, 1994.

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