B. Moghaddam and A. Pentland, "Probabilistic visual learning for object recognition," IEEE Transactions on Pattern Analysis and Machine Intelligence, vol. 19, no. 7, pp. 696-- 710, Juillet 1997.  Home/Search   Document Not in Database   Summary   Related Articles   Check

....learning alone can capture the inner structure of face. Other than defining the boundary of face class v.s. all non face classes, the objective function in [2] makes no attempt to model the target class. One common approach for target class s generative modelling is the eigen face approach [13] [8]. There, PCA is performed to find a reduced dimensional subspace for a holistic modelling of individual faces. In the work of fisherface [1] PCA is performed prior to fisher linear discriminative dimension reduction. This can be thought as a unrefined combination of generative modelling and ....

B. Moghaddam and A. Pentland. Probabilistic visual learning for object recognition. IEEE Transactions on Pattern Analysis and Machine Intelligence, 19(7):696--710, July 1997.

....have been proposed, but are complex to optimize [18] Figure 1: Light field camera array [23] An alternative approach to capturing pose variation is to use an explicit multi view representation which builds a PCA model at several viewpoints. This approach has been used for pure intensity models [2] as well as shape and texture models [t0] A model of inter view variation was also recovered in [t0] and missing views could be reconstructed. However, in this approach views were relatively sparse, and individual features were not matched across views; images were rendered using only texture ....

A.P.B. Moghaddam, "Probabilistic visual learning for object recognition," IEEE Transactions on Pattern Analysis and Machine Intelligence, vol. 19, no. 7, pp. 696 710, 1997.

....achieve lower error rates for face recognition. 1 Motivation and Approach Subspace methods have been applied successfully in numerous visual recognition tasks such as face localization, face recognition, 3D object recognition, and tracking. In particular, Principal Component Analysis (PCA) 20] [13],and Fisher Linear Discriminant (FLD) methods [6] have been applied to face recognition with impressive results. While PCA aims to extract a subspace in which the variance is maximized (or the reconstruction error is minimized) some unwanted variations (due to lighting, facial expressions, ....

B. Moghaddam and A. Pentland. Probabilistic visual learning for object recognition. IEEE PAMI, 19(7):696-710, 1997.

....that are based on either texture, depth, shape and color information, or a combination of them. A comprehensive survey on face detection methods can be found in [1] A probabilistic method based on density estimation in a high dimensional space using an eigenspace decomposition is proposed in [2]. A closely related work is the example based approach in [3] for locating vertically oriented and unoccluded frontal face views at different scales using a number of Gaussian clusters to model the distributions of face and non face patterns. A mixture of linear subspaces has been used to model ....

B. Moghaddam and A. Pentland, "Probabilistic visual learning for object recognition," IEEE Trans. on Pattern Analysis and Machine Intelligence, vol. 19, no. 7, pp. 696--710, July 1997.

....parameters are previously estimated. The method is illustrated on a road sign recognition application. Experiments show significant improvements with respect to standard estimation schemes. 1 Introduction Appearance based representation of objects has recently received considerable attention [10, 9, 20, 4, 16, 14, 17]. One popular approach is the eigenspace representation [10, 9, 20, 4, 11] which allows a substantial dimensionality reduction of the recognition problem. Eigenspace methods involve a reconstruction procedure, which consists in projecting the observation on the training eigenspace. Then, this ....

....recognition application. Experiments show significant improvements with respect to standard estimation schemes. 1 Introduction Appearance based representation of objects has recently received considerable attention [10, 9, 20, 4, 16, 14, 17] One popular approach is the eigenspace representation [10, 9, 20, 4, 11] which allows a substantial dimensionality reduction of the recognition problem. Eigenspace methods involve a reconstruction procedure, which consists in projecting the observation on the training eigenspace. Then, this projection is identified with the closest model of the database. Traditional ....

B. Moghaddam and A. Pentland, "Probabilistic Visual Learning for Object Recognition", IEEE Transaction on Pattern Analysis an Machine Intelligence, vol. 19, no. 7, pp. 696-710, July 1997.

....the many aforementioned variable factors, developing a robust human face detector is a hard task. A comprehensive survey on face detection methods can be found in [1] A probabilistic method based on density estimation in a high dimensional space using an eigenspace decomposition is proposed in [2]. A closely related work is the example based approach in [3] for locating vertically oriented and unoccluded frontal face views at different scales by using a number of Gaussian clusters to model the distributions of face and non face patterns. The color distribution of the face region pixels is ....

B. Moghaddam and A. Pentland, "Probabilistic visual learning for object recognition," IEEE Trans. on Pattern Analysis and Machine Intelligence, vol. 19, no. 7, pp. 696--710, July 1997.

....to a set of class templates; generative methods use a probability density model for each class, and pick the class with the highest likelihood of generating the feature representation; discriminative models compute a discriminant function that directly produces a score for each class. Generative ((Moghaddam and Pentland 1995)) and discriminative models ( Sung and Poggio 1994) Papageorgiou et al. 1998) are often estimated (learned) from training samples. In all of these approaches, the overall performance of the system is largely determined by the quality of the feature extractor and the segmentation technique. The ....

Moghaddam, B., and Pentland, A. (1995). Probabilistic Visual Learning for Object Recognition. MIT Media Lab, Technical Report 326.

....difficult to use standard PCA as a natural component in a probabilistic solution to a computer vision problem. Second, the manifold defined by PCA is necessarily linear. Techniques which address the first of these problems by constructing a density model include Gaussians and mixtures of Gaussians [12]. The second problem has been addressed by considering non linear projective methods such as principal curves and auto encoder neural networks [11] Bregler and Omohundro [5] and Heap and Hogg [9] use mixture representations to try to capture the non linearity of the manifold. However, their model ....

....than the number of data points. For example, it is clearly not feasible to fit an unconstrained mixture of Gaussians directly to the data in the original high dimensional space using maximum likelihood due to the excessive number of parameters in the covariance matrices. Moghaddam and Pentland [12] therefore project the data onto a PCA subspace and then perform density estimation within this lower dimensional space using Gaussian mixtures. While this limits the number of free parameters in the model, the non linearity of the manifold requires the PCA space to have a significantly higher ....

B. Moghaddam and A. Pentland. Probabilistic visual learning for object recognition. IEEE Transactions on Pattern Analysis and Machine Intelligence, 19(7):696--710, 1997.

.... includes nonlinear regression techniques such as classification and regression trees [11] neural networks [12 14] Gaussian process regression [15] support vector classifiers [16] and nearest neighborhood type methods, including eigen space methods that compute distances within subspaces [17, 18]. In contrast, the approach we take here is to use unlabeled data to train a probability density model of the data (or generative model) in an unsupervised fashion. Two common data processing techniques that can be viewed in this way are clustering and linear dimensionality reduction (principal ....

....However, for 4 shapes and 25 transformations, there are 100 distinct clusters in the training set of 200 patterns. Training a mixture model with 100 clusters on 200 patterns would result in severely overfitting the noise. Fig. 8d shows the first 18 principal components, or eigenimages [17, 18], of the training data. It is difficult to imagine how these components can be used to reconstruct the data accurately. C. Clustering faces and facial poses. Fig. 9a shows examples from a training set of 400 jerky images of two people walking across a cluttered background. We trained a TMG with 4 ....

B. Moghaddam and A. Pentland, "Probabilistic visual learning for object recognition," IEEE Transactions on Pattern Analysis and Machine Intelligence, vol. 19, no. 7, pp. 696--710, July 1997.

....The training face images are then mapped onto the eigen space. In the classification phase, the input face image is projected to the same eigen space and classified by an appropriate method. Many different methods have been used for face recognition, such as the Euclidean distance [15] Bayesian [9] and Linear Discriminant Analysis (LDA) 14, 1, 3, 19, 8] Unlike the PCA which encodes information in an orthogonal linear space, the LDA encodes discriminatory information in a linear separable space of which bases are not necessarily orthogonal. Researchers have demonstrated that the LDA based ....

B. Moghaddam and A. Pentland. Probabilistic visual learning for object recognition. PAMI, 19(7):696--710, 1997.

....appearance from new viewpoints given a single image of a person. 11.8 Related Work Statistical models of shape and texture have been widely used for recognition, tracking and synthesis [31, 39, 15, 62] but have tended to only be used with near fronto parallel images. Moghaddam and Pentland [46] describe using view based eigenface models to represent a wide variety of viewpoints. Our work is similar to this, but by including shape variation (rather than the rigid eigen patches) we require fewer models and can obtain better reconstructions with fewer model modes. Kruger [35] and Wiskott ....

B. Moghaddam and A. Pentland. Probabilistic visual learning for object recognition. IEEE Transactions on Pattern Analysis and Machine Intelligence, 19(7):696-710, 1997.

....1 , e k ) T y(I ) The object om with minimum distance dm between its manifold and g(I ) is assumed to be the object in question: dm = min o i min # j #g(I ) g o i ,# j #. 1) We extend Nayar and Murase s concept of manifolds by introducing probability densities in feature space [4, 19]. Let us denote by p(g o i ,# j ) the likelihood of obtaining the feature vector g after processing an image of object o i with pose parameters # j . The likelihood is estimated from a set of sample images with fixed o i ,# j . The samples capture the inaccuracies in the parameters # such as ....

Moghaddam, B., Pentland, A.: Probabilistic visual learning for object recognition. IEEE Trans. Pattern Anal. Mach. Intell. 19, 696--710 (1997).

....1 , e k ) T y(I) The object o m with minimum distance dm between its manifold and g(I) is assumed to be the object in question: dm = min o i min # j #g(I) g o i ,# j #. 1) We extend Nayar and Murase s concept of manifolds by introducing probability densities in feature space [3, 18]. Let us denote by p(g o i , # j ) the likelihood of obtaining the feature vector g after processing an image of object o i with pose parameters # j . The likelihood is estimated from a set of sample images with fixed o i , # j . The samples capture the inaccuracies in the parameters # such as ....

B. Moghaddam and A. Pentland. Probabilistic Visual Learning for Object Recognition. IEEE Transactions on Pattern Analysis and Machine Intelligence, 19(7):696--710, July 1997.

....i;j and g y is assumed to be the object in question. Note that eq. 1) gives us both: an object hypothesis and a pose estimation. dm = min i min j kg y Gamma g i;j k: 1) It will prove useful to extend Nayar and Murase s concept of manifolds by introducing probability densities in eigenspace [4, 16]. Let us denote by 1 In order to simplify notation we assume X having zero mean. 2 Sufficient in the sense of sufficient for disambiguating various objects. Quantitatively we demand P k i=1 i =T race(Q) threshold. 5 p(gjo i ; j ) the likelihood of ending up at point g in the eigenspace ....

B. Moghaddam and A. Pentland. Probabilistic Visual Learning for Object Recognition. IEEE Transactions on Pattern Analysis and Machine Intelligence, 19(7):696--710, July 1997.

....kernel methods with Eigenface methods on two benchmarks. Empirical results show that Kernel PCA outperforms Eigenface method in face recognition. 1. MOTIVATION AND APPROACH Subspace methods have been applied successfully in applications such as face recognition using Eigenfaces (or PCA face) 8] [3], face detection [3] object recognition [4] and tracking [1] Representations such as PCA encodes the pattern information based on second order dependencies, i.e. pixelwise covariance among the pixels, and are insensitive to the dependencies of multiple (more than two) pixels in the patterns. ....

....Eigenface methods on two benchmarks. Empirical results show that Kernel PCA outperforms Eigenface method in face recognition. 1. MOTIVATION AND APPROACH Subspace methods have been applied successfully in applications such as face recognition using Eigenfaces (or PCA face) 8] 3] face detection [3], object recognition [4] and tracking [1] Representations such as PCA encodes the pattern information based on second order dependencies, i.e. pixelwise covariance among the pixels, and are insensitive to the dependencies of multiple (more than two) pixels in the patterns. Since the ....

B. Moghaddam and A. Pentland. Probabilistic visual learning for object recognition. IEEE Transactions on Pattern Analysis and Machine Intelligence, 19(7):696-710, 1997.

....maximal discrimination between positive and negative examples of faces. They use a family of discrete Markov processes to model the face and background patterns and estimate the density functions. Detection of a face is based on the likelihood ratio computed during training. Moghaddam and Pentland [10] propose a probabilistic method that is based on density estimation in a high dimensional space using an eigenspace decomposition. In [16] Rowley et al. use an ensemble of neural networks to learn face and nonface patterns for face detection. Schneiderman et al. describe a probabilistic method ....

....as follows. First, we introduce a projection method that performs better than PCA. Consequently, the classi cation in the linear subspace is better. Second, we apply a mixture model such that the linear subspaces can better capture the variations of face patterns. Although some methods [10] [19] have applied mixture models, they use PCA for projection which is not optimal for classi cation in subspaces. On the other hand, it is not clear how SVM performs in face detection since the study in [11] has applied SVM on a rather small test set with 136 faces. It will be of great interest ....

B. Moghaddam and A. Pentland. Probabilistic visual learning for object recognition. IEEE Transactions on Pattern Analysis and Machine Intelligence, 19(7):696{ 710, 1997.

No context found.

B. Moghaddam and A. Pentland, "Probabilistic visual learning for object recognition," IEEE Transactions on Pattern Analysis and Machine Intelligence, vol. 19, no. 7, pp. 696-- 710, Juillet 1997.

No context found.

B. Moghaddam and A. Pentland, "Probabilistic visual learning for object recognition," IEEE Transactions on Pattern Analysis and Machine Intelligence, vol. 19, no. 7, pp. 696-- 710, Juillet 1997.

No context found.

B. Moghaddam and A. Pentland. Probabilistic visual learning for object recognition. IEEE Transactions on Pattern Analysis and Machine Intelligence, 19(7):696--710, 1997.

No context found.

B. Moghaddam and A. Pentland. Probabilistic visual learning for object recognition. IEEE Transactions on Pattern Analysis and Machine Intelligence, 19(7):696-710, 1997.

No context found.

B. Moghaddam and A.P. Pentland. Probabilistic visual learning for object recognition. IEEE Trans. on PAMI , 19(7):696--710, July 1997.

No context found.

B. Moghaddam and A. Pentland. Probabilistic visual learning for object recognition. IEEE Transactions on Pattern Analysis and Machine Intelligence, 19(7):696--710, 1997.

No context found.

B. Moghaddam and A.P. Pentland. Probabilistic visual learning for object recognition. IEEE Trans. on PAMI , 19(7):696--710, July 1997.

No context found.

Moghaddam B. and Pentland A. #1996# Probabilistic Visual Learning for Object Recognition, in Nayar S. and Poggio T. #eds.# "Early Visual Learning", Oxford University Press.

No context found.

MOGHADDAM B. AND PENTLAND A. 1996 Probabilistic Visual Learning for Object Recognition, in Nayar S. and Poggio T. (eds.), Early Visual Learning, Oxford University Press.

First 50 documents

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