Nonparametric feature selection

Patrick, Edward A.; Fischer, Frederic P.

doi:10.1109/tit.1969.1054354

Cited by 48 publications

(14 citation statements)

References 3 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…These measures were designed to capture such differences without assuming the distribution or any prior knowledge of the samples. Therefore, statistical distance measures (that are also computationally more complicated), such as Mahalanobis [51], Kolmogorov [1], Bhattacharyya [7], Bayesian distance [8], Chernoff [12], Matsusita [52], and divergence [40], [46], and those utilized in pattern recognition [50], [57], have not been used in our experiment.…”

Section: A Quantizationmentioning

confidence: 99%

Texture analysis of SAR sea ice imagery using gray level co-occurrence matrices

Soh

Tsatsoulis

1999

IEEE Trans. Geosci. Remote Sensing

1,058

498

View full text Add to dashboard Cite

Abstract-This paper presents a preliminary study for mapping sea ice patterns (texture) with 100-m ERS-1 synthetic aperture radar (SAR) imagery. We used gray-level co-occurrence matrices (GLCM) to quantitatively evaluate textural parameters and representations and to determine which parameter values and representations are best for mapping sea ice texture. We conducted experiments on the quantization levels of the image and the displacement and orientation values of the GLCM by examining the effects textural descriptors such as entropy have in the representation of different sea ice textures. We showed that a complete gray-level representation of the image is not necessary for texture mapping, an eight-level quantization representation is undesirable for textural representation, and the displacement factor in texture measurements is more important than orientation. In addition, we developed three GLCM implementations and evaluated them by a supervised Bayesian classifier on sea ice textural contexts. This experiment concludes that the best GLCM implementation in representing sea ice texture is one that utilizes a range of displacement values such that both microtextures and macrotextures of sea ice can be adequately captured. These findings define the quantization, displacement, and orientation values that are the best for SAR sea ice texture analysis using GLCM.

show abstract

Section: A Quantizationmentioning

confidence: 99%

Texture analysis of SAR sea ice imagery using gray level co-occurrence matrices

Soh

Tsatsoulis

1999

IEEE Trans. Geosci. Remote Sensing

1,058

498

View full text Add to dashboard Cite

show abstract

“…Several supervised linear dimensionality reduction methods exist in the literature. We can group these methods into three broad categories: those that separate likelihood functions according to some distance or divergence [38]- [44], those that try to make the probability of the labels given the measurements and the probability of the labels given the dimensionality-reduced measurements equal [45]- [50] and those that attempt to minimize a specific classification or regression objective [12], [51]- [54].…”

Section: A Relationship To Prior Workmentioning

confidence: 99%

“…The method of [38], like FDA, maximally separates the clusters of the different labels but does not make the strong Gaussian assumption. Instead, it performs kernel density estimation of the likelihoods and separates those estimates.…”

Section: A Relationship To Prior Workmentioning

confidence: 99%

“…Like [38] and [44], the method of [49] also estimates probability density functions for use in the criterion for linear dimensionality reduction. The particular criterion, however, is based on the idea that the dimensionality reduction mapping should be such that the probability of the class labels conditioned on the unreduced measurements equal the probability conditioned on the reduced measurements.…”

Section: A Relationship To Prior Workmentioning

confidence: 99%

See 1 more Smart Citation

Linear Dimensionality Reduction for Margin-Based Classification: High-Dimensional Data and Sensor Networks

Varshney

Willsky

2011

IEEE Trans. Signal Process.

View full text Add to dashboard Cite

Abstract-Low-dimensional statistics of measurements play an important role in detection problems, including those encountered in sensor networks. In this work, we focus on learning low-dimensional linear statistics of high-dimensional measurement data along with decision rules defined in the low-dimensional space in the case when the probability density of the measurements and class labels is not given, but a training set of samples from this distribution is given. We pose a joint optimization problem for linear dimensionality reduction and margin-based classification, and develop a coordinate descent algorithm on the Stiefel manifold for its solution. Although the coordinate descent is not guaranteed to find the globally optimal solution, crucially, its alternating structure enables us to extend it for sensor networks with a message-passing approach requiring little communication. Linear dimensionality reduction prevents overfitting when learning from finite training data. In the sensor network setting, dimensionality reduction not only prevents overfitting, but also reduces power consumption due to communication. The learned reduced-dimensional space and decision rule is shown to be consistent and its Rademacher complexity is characterized. Experimental results are presented for a variety of datasets, including those from existing sensor networks, demonstrating the potential of our methodology in comparison with other dimensionality reduction approaches.

show abstract

“…Patrick and Fischer [2] represent an early example of using kernel density estimators to project high-dimensional measurements to low-dimensional representations, albeit in a different context. Our approach combines the flexibility of nonparametric estimates with information-preserving subspaces in a dependency scenario.…”

Section: Introductionmentioning

confidence: 99%

Estimating Dependency and Significance for High-Dimensional Data

Siracusa

Tieu

Ihler³

et al.

Proceedings. (ICASSP '05). IEEE International Conference on Acoustics, Speech, and Signal Processing, 2005.

View full text Add to dashboard Cite

Understanding the dependency structure of a set of variables is a key component in various signal processing applications which involve data association. The simple task of detecting whether any dependency exists is particularly difficult when models of the data are unknown or difficult to characterize because of high-dimensional measurements. We review the use of nonparametric tests for characterizing dependency and how to carry out these tests with highdimensional observations. In addition we present a method to assess the significance of the tests.

show abstract

Nonparametric feature selection

Cited by 48 publications

References 3 publications

Texture analysis of SAR sea ice imagery using gray level co-occurrence matrices

Texture analysis of SAR sea ice imagery using gray level co-occurrence matrices

Linear Dimensionality Reduction for Margin-Based Classification: High-Dimensional Data and Sensor Networks

Estimating Dependency and Significance for High-Dimensional Data

Contact Info

Product

Resources

About