Kevin R. Vixie

Abel inversion using total-variation regularization

In the case of radiography of a cylindrically symmetric object, the Abel transform is useful for describing the tomographic measurement operator. The inverse of this operator is unbounded, so regularization is required for the computation of satisfactory inversions. We introduce the use of the total variation seminorm for this purpose, and prove the existence and uniqueness of solutions of the corresponding variational problem. We illustrate the effectiveness of the total-variation regularization with an example and comparison with the unregularized inverse and the H1 regularized inverse.

Incorporating invariants in Mahalanobis distance based classifiers: application to face recognition

We present a technique for combining prior knowledge about transformations that should be ignored with a covariance matrix estimated from training data to make an improved Mahalanobis distance classifier. Modern classification problems often involve objects represented by high-dimensional vectors or images (for example, sampled speech or human faces). The complex statistical structure of these representations is often difficult to infer from the relatively limited training data sets that are available in practice. Thus, we wish to efficiently utilize any available a priori information, such as transformations or the representations with respect to which the associated objects are known to retain the same classification (for example, spatial shifts of an image of a handwritten digit do not alter the identity of the digit). These transformations, which are often relatively simple in the space of the underlying objects, are usually nonlinear in the space of the object representation, making their inclusion within the framework of a standard statistical classifier difficult. Motivated by prior work of Simard et al. (1998; 2000), we have constructed a new classifier which combines statistical information from training data and linear approximations to known invariance transformations. When tested on a face recognition task, performance was found to exceed by a significant margin that of the best algorithm in a reference software distribution.

Abel inversion using total variation regularization: applications

We apply total-variation (TV) regularization methods to Abel inversion tomography. Inversions are performed using the fixed-point iteration method and the regularization parameter is chosen such that the resulting data fidelity approximates the known or estimated statistical character of the noisy data. Five one-dimensional (1D) examples illustrate the favorable characteristics of TV-regularized solutions: noise suppression and density discontinuity preservation. Experimental and simulated examples from X-ray radiography also illustrate limitations due to a linear projection approximation. TV-regularized inversions are shown to be superior to squared gradient (Tikhonov) regularized inversions for objects with density discontinuities. We also introduce an adaptive TV method that utilizes a modified discrete gradient operator acting only apart from data-determined density discontinuities. This method provides improved density level preservation relative to the basic TV method.

L1TV Computes the Flat Norm for Boundaries

We show that the recently introduced L1TV functional can be used to explicitly compute the flat norm for codimension one boundaries. Furthermore, using L1TV, we also obtain the flat norm decomposition. Conversely, using the flat norm as the precise generalization of L1TV functional, we obtain a method for denoising nonboundary or higher codimension sets. The flat norm decomposition of differences can made to depend on scale using the flat norm with scale which we define in direct analogy to the L1TV functional. We illustrate the results and implications with examples and figures.

Simplicial flat norm with scale

We study the multiscale simplicial flat norm (MSFN) problem, which computes flat norm at various scales of sets defined as oriented subcomplexes of finite simplicial complexes in arbitrary dimensions. We show that the multiscale simplicial flat norm is NP-complete when homology is defined over integers. We cast the multiscale simplicial flat norm as an instance of integer linear optimization. Following recent results on related problems, the multiscale simplicial flat norm integer program can be solved in polynomial time by solving its linear programming relaxation, when the simplicial complex satisfies a simple topological condition (absence of relative torsion). Our most significant contribution is the simplicial deformation theorem, which states that one may approximate a general current with a simplicial current while bounding the expansion of its mass. We present explicit bounds on the quality of this approximation, which indicate that the simplicial current gets closer to the original current as we make the simplicial complex finer. The multiscale simplicial flat norm opens up the possibilities of using flat norm to denoise or extract scale information of large data sets in arbitrary dimensions. On the other hand, it allows one to employ the large body of algorithmic results on simplicial complexes to address more general problems related to currents.

Classification Modulo Invariance, With Application to Face Recognition

This article presents techniques for constructing classifiers that combine statistical information from training data with tangent approximations to known transformations; it demonstrates the techniques by applying them to a face recognition task. Our approach is to build Bayes classifiers with approximate class-conditional probability densities for measured data. The high dimension of the measurements in modern classification problems such as speech or image recognition makes inferring probability densities from feasibly sized training datasets difficult. We address the difficulty by imposing severely simplifying assumptions and exploiting a priori information about transformations to which classification should be invariant. For the face recognition task, we used a five-parameter group of such transformations consisting of rotation, shifts, and scalings. On the face recognition task, a classifier based on our techniques has an error rate that is 20% lower than that of the best algorithm in a reference software distribution.

The bispectral aliasing test: a clarification and some key examples

Controversy regarding the correctness of a test for aliasing proposed by Hinich and Wolinsky (1988) has been surprisingly long-lived. Two factors have prolonged this controversy. One factor is the presence of deep-seated intuitions that such a test is fundamentally impossible. Perhaps the most compelling objection is that, given a set of discrete-time samples, one can construct an unaliased continuous-time series which exactly fits those samples. Therefore, the samples alone can not show that the original time series was aliased. The second factor prolonging the debate has been an inability of its proponents to unseat those objections. In fact, as is shown here, all objections can be met and the test as stated is correct. In particular, the role of stationarity as prior knowledge in addition to the sample values turns out to be crucial. Under certain conditions, including those addressed by the bispectral aliasing test, the continuous-time signals reconstructed from aliased samples are nonstationary. Therefore detecting aliasing in (at least some) stationary continuous-time processes both makes sense and can be done. The merits of the bispectral test for practical use are addressed, but our primary concern here is its theoretical soundness.

Invariant template matching with tangent vectors

Template matching is the search for a known object, represented by a template image, at an arbitrary location within a larger image. The local measure of match is often desired to be invariant to certain transforms, such as rotation and dilation, of the template. Although a variety of solutions have been proposed, most are designed to provide invariance to a specific transform or set of transforms, and often involve significant computational demands. When invariance to “small” transformations of the template (e.g., rotation by a small angle) is sufficient, local linear approximations to these transforms may be used to allow template matching with invariance to arbitrary transforms, without significantly increased computational requirements.

Flat norm decomposition of integral currents

Currents represent generalized surfaces studied in geometric measure theory. They range from relatively tame integral currents representing oriented compact manifolds with boundary and integer multiplicities, to arbitrary elements of the dual space of differential forms. The flat norm provides a natural distance in the space of currents, and works by decomposing a

Nonasymptotic Densities for Shape Reconstruction

d