Evaluating multi-class multiple-instance learning for image categorization

Xinyu Xu, Baoxin Li

Research output: Chapter in Book/Report/Conference proceedingConference contribution

4 Scopus citations

Abstract

Automatic image categorization is a challenging computer vision problem, to which Multiple-instance Learning (MIL) has emerged as a promising approach. Typical current MIL schemes rely on binary one-versus-all classification, even for inherently multi-class problems. There are a few drawbacks with binary MIL when applied to a multi-class classification problem. This paper describes Multi-class Multiple-Instance Learning (McMIL) to image categorization that bypasses the necessity of constructing a series of binary classifiers. We analyze McMIL in depth to show why it is advantageous over binary MIL when strong target concept overlaps exist among the classes. We systematically val-ate McMIL using two challenging image databases, and compare it with state-of-the-art binary MIL approaches. The McMIL achieves competitive classification accuracy, robustness to labeling noise, and effectiveness in capturing the target concepts using smaller amount of training data. We show that the learned target concepts from McMIL conform to human interpretation of the images.

Original languageEnglish (US)
Title of host publicationComputer Vision - ACCV 2007 - 8th Asian Conference on Computer Vision, Proceedings
PublisherSpringer Verlag
Pages155-165
Number of pages11
EditionPART 2
ISBN (Print)9783540763895
DOIs
StatePublished - Jan 1 2007
Event8th Asian Conference on Computer Vision, ACCV 2007 - Tokyo, Japan
Duration: Nov 18 2007Nov 22 2007

Publication series

NameLecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)
NumberPART 2
Volume4844 LNCS
ISSN (Print)0302-9743
ISSN (Electronic)1611-3349

Other

Other8th Asian Conference on Computer Vision, ACCV 2007
CountryJapan
CityTokyo
Period11/18/0711/22/07

Keywords

  • Image categorization
  • Multi-class multiple-instance learning

ASJC Scopus subject areas

  • Theoretical Computer Science
  • Computer Science(all)

Fingerprint Dive into the research topics of 'Evaluating multi-class multiple-instance learning for image categorization'. Together they form a unique fingerprint.

  • Cite this

    Xu, X., & Li, B. (2007). Evaluating multi-class multiple-instance learning for image categorization. In Computer Vision - ACCV 2007 - 8th Asian Conference on Computer Vision, Proceedings (PART 2 ed., pp. 155-165). (Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics); Vol. 4844 LNCS, No. PART 2). Springer Verlag. https://doi.org/10.1007/978-3-540-76390-1_16