Evaluating multi-class multiple-instance learning for image categorization

Xinyu Xu, Baoxin Li

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

4 Citations (Scopus)

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 publicationLecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)
Pages155-165
Number of pages11
Volume4844 LNCS
EditionPART 2
StatePublished - 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)03029743
ISSN (Electronic)16113349

Other

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

Fingerprint

Multi-class
Categorization
Learning
Binary
Labeling
Computer vision
Classifiers
Target
Competitive Learning
Multi-class Classification
Image Database
Classification Problems
Computer Vision
Overlap
Noise
Classifier
Robustness
Databases
Series

Keywords

  • Image categorization
  • Multi-class multiple-instance learning

ASJC Scopus subject areas

  • Computer Science(all)
  • Biochemistry, Genetics and Molecular Biology(all)
  • Theoretical Computer Science

Cite this

Xu, X., & Li, B. (2007). Evaluating multi-class multiple-instance learning for image categorization. In Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics) (PART 2 ed., Vol. 4844 LNCS, 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).

Evaluating multi-class multiple-instance learning for image categorization. / Xu, Xinyu; Li, Baoxin.

Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics). Vol. 4844 LNCS PART 2. ed. 2007. p. 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).

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

Xu, X & Li, B 2007, Evaluating multi-class multiple-instance learning for image categorization. in Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics). PART 2 edn, vol. 4844 LNCS, Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics), no. PART 2, vol. 4844 LNCS, pp. 155-165, 8th Asian Conference on Computer Vision, ACCV 2007, Tokyo, Japan, 11/18/07.
Xu X, Li B. Evaluating multi-class multiple-instance learning for image categorization. In Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics). PART 2 ed. Vol. 4844 LNCS. 2007. p. 155-165. (Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics); PART 2).
Xu, Xinyu ; Li, Baoxin. / Evaluating multi-class multiple-instance learning for image categorization. Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics). Vol. 4844 LNCS PART 2. ed. 2007. pp. 155-165 (Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics); PART 2).
@inproceedings{33801cdbda0d437a8d314686cdd850e4,
title = "Evaluating multi-class multiple-instance learning for image categorization",
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.",
keywords = "Image categorization, Multi-class multiple-instance learning",
author = "Xinyu Xu and Baoxin Li",
year = "2007",
language = "English (US)",
isbn = "9783540763895",
volume = "4844 LNCS",
series = "Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)",
number = "PART 2",
pages = "155--165",
booktitle = "Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)",
edition = "PART 2",

}

TY - GEN

T1 - Evaluating multi-class multiple-instance learning for image categorization

AU - Xu, Xinyu

AU - Li, Baoxin

PY - 2007

Y1 - 2007

N2 - 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.

AB - 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.

KW - Image categorization

KW - Multi-class multiple-instance learning

UR - http://www.scopus.com/inward/record.url?scp=38349018853&partnerID=8YFLogxK

UR - http://www.scopus.com/inward/citedby.url?scp=38349018853&partnerID=8YFLogxK

M3 - Conference contribution

AN - SCOPUS:38349018853

SN - 9783540763895

VL - 4844 LNCS

T3 - Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)

SP - 155

EP - 165

BT - Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)

ER -