Mixtures of Normalized Linear Projections

Ahmed Fawzi Otoom, Oscar Perez Concha, Hatice Gunes, Massimo Piccardi

Research output: Chapter in Book/Report/Conference proceedingChapterpeer-review


High dimensional spaces pose a challenge to any classification task. In fact, these spaces contain much redundancy and it becomes crucial to reduce the dimensionality of the data to improve analysis, density modeling, and classification. In this paper, we present a method for dimensionality reduction in mixture models and its use in classification. For each component of the mixture, the data are projected by a linear transformation onto a lower-dimensional space. Subsequently, the projection matrices and the densities in such compressed spaces are learned by means of an Expectation Maximization (EM) algorithm. However, two main issues arise as a result of implementing this approach, namely: 1) the scale of the densities can be different across the mixture components and 2) a singularity problem may occur. We suggest solutions to these problems and validate the proposed method on three image data sets from the UCI Machine Learning Repository. The classification performance is compared with that of a mixture of probabilistic principal component analysers (MPPCA). Across the three data sets, our accuracy always compares favourably, with improvements ranging from 2.5% to 35.4%.

Original languageEnglish
Title of host publicationAdvanced Concepts for Intelligent Vision Systems
Subtitle of host publication11th International Conference, ACIVS 2009, Bordeaux, France, September 28–October 2, 2009. Proceedings
EditorsJacques Blanc-Talon, Wilfried Philips, Dan Popescu, Paul Scheunders
Place of PublicationBerlin
PublisherSpringer, Springer Nature
Number of pages11
ISBN (Electronic)9783642046971
ISBN (Print)3642046967, 9783642046964
Publication statusPublished - 2009
Externally publishedYes
Event11th International Conference on Advanced Concepts for Intelligent Vision Systems, ACIVS 2009 - Bordeaux, France
Duration: 28 Sept 20092 Oct 2009

Publication series

NameLecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)
PublisherSpringer Berlin Heidelberg
ISSN (Print)0302-9743
ISSN (Electronic)1611-3349


Other11th International Conference on Advanced Concepts for Intelligent Vision Systems, ACIVS 2009


Dive into the research topics of 'Mixtures of Normalized Linear Projections'. Together they form a unique fingerprint.

Cite this