We present a variational integration of nonlinear shape statistics into a Mumford-Shah based segmentation process. The nonlinear statistics are derived from a set of training silhouettes by a novel method of density estimation which can be considered as an extension of kernel PCA to a stochastic framework.

High information redundancy and strong correlations in face images result in inefficiencies when such images are used directly in recognition tasks. In this paper, Discrete Cosine Transforms (DCTs) are used to reduce image information redundancy because only a subset of the transform coefficients are necessary to preserve the most important facial features, such as hair outline, eyes and mouth. We demonstrate experimentally that when DCT coefficients are fed into a backpropagation neural network for classification, high recognition rates can be achieved using only a small proportion (0.19%) of available transform components. This makes DCT-based face recognition more than two orders of magnitude faster than other approaches.

High information redundancy and correlation in face images result in ineciencies when such images are used directly for recognition. In this paper, discrete cosine transforms are used to reduce image information redundancy because only a subset of the transform coecients are necessary to preserve the most important facial features such as hair outline, eyes and mouth. We demonstrate experimentally that when DCT coecients are fed into a backpropagation neural network for classi cation, a high recognition rate can be achieved by using a very small proportion of transform coecients. This makes DCT-based face recognition much faster than other approaches. Key words: Face recognition, neural networks, feature extraction, discrete cosine transform. 1 Introduction High information redundancy present in face images results in ineciencies when these images are used directly for recognition, identi cation and classi cation. Typically one builds a computational model to transform pixel i...

We present a novel approach for representing shape knowledge in terms of example views of 3D objects. Typically, such data sets exhibit a highly nonlinear structure with distinct clusters in the shape vector space, preventing the usual encoding by linear principal component analysis (PCA). For this reason, we propose a nonlinear Mercer kernel PCA scheme which takes into account both the projection distance and the within-subspace distance in a high-dimensional feature space. The comparison of our approach with supervised mixture models indicates that the statistics of example views of distinct 3D objects can fairly well be learned and represented in a completely unsupervised way.

An integral part of any audio-visual speech processing (AVSP) system, is the front-end visual system that detects facial features (e.g. eyes and mouth) pertinent to the task of visual speech processing. The ability of this front-end system to not only locate, but give a confidence measure that the facial feature is present in the image, directly affects the ability of any subsequent post-processing task such as speech or speaker recognition. With these issues in mind, this paper presents a framework for a facial feature detection system suitable for use in an AVSP system, but whose basic framework is useful for any application requiring frontal facial feature detection. A novel approach for facial feature detection is presented based on an appearance paradigm. This approach, based on intra-class unsupervised clustering and discriminant analysis, displays improved detection performance over conventional techniques.

Principal curves, like principal components, are a tool used in multivariate analysis for ends like feature extraction. Defined in their original form, principal curves need not exist for general distributions. The existence of principal curves with bounded length for any distribution that satisfies some minimal regularity conditions has been shown. We define principal curves with bounded turn, show that they exist, and present a learning algorithm for them. Principal components are a special case of such curves when the turn is zero.

Principal Components Analysis (PCA) is traditionally a linear technique for projecting multidimensional data onto lower dimensional subspaces with minimal loss of variance. However, there are several applications where the data lie in a lower dimensional subspace that is not linear; in these cases linear PCA is not the optimal method to recover this subspace and thus account for the largest proportion of variance in the data. Nonlinear

Principal Component Analysis (PCA) approaches to image recognition are data dependent and computationally expensive. To classify unknown images they need to match the nearest neighbour in the stored database of extracted image features. In this paper, Discrete Cosine Transforms (DCTs) are used to reduce the dimensionality of image space by truncating high frequency DCT components. The remaining coefficients are fed into a neural network for classification. Because only a small number of low frequency DCT components are necessary to preserve the most important facial features such as hair outline, eyes and mouth, our DCT-based image recognition system is much faster than other approaches. 1. INTRODUCTION A common objective in face recognition is to find a good way of representing face information. A key step in developing a good representation is to expose the constraints and remove the redundancies contained in pixel images of faces. A well-known and widely used statistical technique...

Abstract. Concepts from elasticity are applied to analyze modes of variation on shapes in two and three dimensions. This approach represents a physically motivated alternative to shape statistics on a Riemannian shape space, and it robustly treats strong nonlinear geometric variations of the input shapes. To compute a shape average, all input shapes are elastically deformed into the same configuration. That configuration which minimizes the total elastic deformation energy is defined as the average shape. Each of the deformations from one of the shapes onto the shape average induces a boundary stress. Small amplitude stimulation of these stresses leads to displacements which reflect the impact of every single input shape on the average. To extract the dominant modes of variation, a PCA is performed on this set of displacements. To make the approach computationally tractable, a relaxed formulation is proposed, and sharp contours are approximated via phase fields. For the spatial discretization of the resulting model, piecewise multilinear finite elements are applied. Applications in 2D and in 3D demonstrate the qualitative properties of the presented approach. 1

In many vision problems, the observed data lies in a nonlinear manifold in a high-dimensional space. This paper presents a generic modelling scheme to characterize the nonlinear structure of the manifold and to learn its multimodal distribution. Our approach represents the data as a linear combination of parameterized local components, where the statistics of the component parameterization describe the nonlinear structure of the manifold. The components are adaptively selected from the training data through a progressive density approximation procedure, which leads to the maximum likelihood estimate of the underlying density. We show results on both synthetic and real training sets, and demonstrate that the proposed scheme has the ability to reveal important structures of the data.