Learning a Mahalanobis distance metric for data clustering and classification

Xiang, Shiming; Nie, Feiping; Zhang, Changshui

doi:10.1016/j.patcog.2008.05.018

Cited by 514 publications

(262 citation statements)

References 12 publications

Supporting

Mentioning

243

Contrasting

Unclassified

Order By: Relevance

“…Xiang et al [5] propose to learn a quadratic metric referred to as the discriminant metric learning (DML). The subtle difference between DML and LDA is, DML tries to optimize the average distance between data, while LDA tries to optimize the average distance between data and data mean; they are closely connected but not the same thing.…”

Section: The Translation Invariant Solutionsmentioning

confidence: 99%

“…We have to point out that the translation-invariant solution to DML [5] is exactly the same as the transform domain metric learning (TDML) [1].…”

Section: The Translation Invariant Solutionsmentioning

confidence: 99%

“…Despite that Fukunaga and Flick published their pioneer work in 1984 [2], the term distance metric learning is due to Xing et al [3], and learning a distance metric from data has been actively studied since then [1,3,4,5].…”

Section: Introductionmentioning

confidence: 99%

See 2 more Smart Citations

The Translation Invariant Solution to Quadratic Metric Learning

Sun

Feng

Shen³

2015

Proceedings of the 2015 2nd International Workshop on Materials Engineering and Computer Sciences

View full text Add to dashboard Cite

Abstract. Metric learning has drawn great interests for the last decade in the field of computer vision and machine learning. In this paper, we address the importance of the translation invariance (TI) of a metric. Intuitively, translation invariance should be a fundamental requirement for any reasonable image metric, but few metric learning or subspace methods are aware of the TI property when dealing with images. We propose to solve the quadratic metric learning problem in the transform domain based on the result of [1]. The solution is guaranteed to be translation invariant. The TI assumption simplifies the optimization problems considerably. Specifically, it reduces the number of optimization parameters from of O(n2) to O(n); it turns the semi-definite constraint on a matrix to a bound constraint on a function; it applies to multi-dimensional data without having to be stacked to vectors. Experimental results show that the TI solutions are generally on par with the non-TI solutions. Specifically, the TI solutions usually favor small sample size and small reduction dimension. The framework proposed for quadratic metric learning and linear subspace method seems quite promising.

show abstract

Section: The Translation Invariant Solutionsmentioning

confidence: 99%

“…We have to point out that the translation-invariant solution to DML [5] is exactly the same as the transform domain metric learning (TDML) [1].…”

Section: The Translation Invariant Solutionsmentioning

confidence: 99%

See 1 more Smart Citation

The Translation Invariant Solution to Quadratic Metric Learning

Sun

Feng

Shen³

2015

Proceedings of the 2015 2nd International Workshop on Materials Engineering and Computer Sciences

View full text Add to dashboard Cite

show abstract

“…All P (ω 1 |v i (t)) (1 ≤ i ≤ I) will constitute a new feature vector v. According to the extraction method, all features have similar accuracy for target description. So we make a classification with the combination of features using the Mahalanobis distance [29]. The new feature vector of training samples is written as V = {v 1 , v 2 , . .…”

Section: Discriminator Based On Bayesian Decision Rulementioning

confidence: 99%

Sparse Time-Frequency Representation Based Feature Extraction Method for Landmine Discrimination

Wang¹,

Song²,

Tian³

et al. 2013

PIER

View full text Add to dashboard Cite

Abstract-Low-frequency ultra-wideband synthetic aperture radar is a promising technology for landmine detection. According to the scattering characteristics of body-of-revolution (BOR) along with azimuth angles, a discriminator based on Bayesian decision rule is proposed, which uses sequential features, i.e., double-hump distance. First, the algorithm estimates the target scatterings in all azimuth angles based on regions of interest. Second, sequential aspect features are extracted by sparse time-frequency representation. Third, the distributions of features are obtained by training samples, and then the posterior probability of landmine class is computed as an input to the classifier adopting Mahalanobis distance. The experimental results indicate that the proposed algorithm is effective in BOR target discrimination.

show abstract

“…To measure the similarity of the Hu moments we used Mahalanobis distance [10]. This gives us the distance between the mean of the training set and the input frame.…”

Section: B) Statistical Comparisons Of Strokesmentioning

confidence: 99%

Recognition of tennis strokes using key postures

Connaghan

Conaire

Kelly

et al. 2010

IET Irish Signals and Systems Conference (ISSC 2010)

View full text Add to dashboard Cite

-In this paper we describe an approach for automatic recognition of tennis strokes using a single low cost camera. Professional tennis is played at high speed so the ability to classify tennis strokes on camera is hindered by the rapid movement of the players. We have developed an accurate recognition system which can automatically index tennis strokes from video footage. We aim to evolve this system so that meta data, such as time codes and descriptions of the strokes played, can be automatically indexed for a training session or a match. This level of indexing would provide an excellent foundation for the development of next generation sports coaching systems. The aim of this paper is to coarsely classify the main stokes played in tennis, i.e. a serve, forehand or backhand. The proposed approach is evaluated against a real-world dataset, obtained from elite players in a competitive training match.

show abstract

Learning a Mahalanobis distance metric for data clustering and classification

Cited by 514 publications

References 12 publications

The Translation Invariant Solution to Quadratic Metric Learning

The Translation Invariant Solution to Quadratic Metric Learning

Sparse Time-Frequency Representation Based Feature Extraction Method for Landmine Discrimination

Recognition of tennis strokes using key postures

Contact Info

Product

Resources

About