Combining holistic and object-based approaches for scene classification

Zenghai Chen, Zheru Chi, Hong Fu, Dagan Feng

Research output: Chapter in book / Conference proceedingConference article published in proceeding or bookAcademic researchpeer-review

2 Citations (Scopus)

Abstract

There are two main approaches for scene classification: holistic and object-based. Holistic approach is good at representing scenes with simple content. However, since it does not take into account the internal object relationship, holistic approach does not well characterize complex scenes with multiple objects. By contrast, object-based approach estimates the scene class by analyzing the object co-occurrence information, as a result of which it is advantageous in characterizing scenes with complex content. But object-based approach is not good at classifying simple scenes. In this paper, we combine holistic and object-based approaches for scene classification. The proposed combinatory approach is able to take advantages of the two approaches. Several state-of-the-art holistic and object-based approaches are compared. The experiments conducted on a widely-used scene dataset demonstrate the superiors performance of the combinatory approach.
Original languageEnglish
Title of host publicationProceedings - 2012 5th International Symposium on Computational Intelligence and Design, ISCID 2012
Pages65-68
Number of pages4
Volume1
DOIs
Publication statusPublished - 1 Dec 2012
Event2012 5th International Symposium on Computational Intelligence and Design, ISCID 2012 - Hangzhou, China
Duration: 28 Oct 201229 Oct 2012

Conference

Conference2012 5th International Symposium on Computational Intelligence and Design, ISCID 2012
Country/TerritoryChina
CityHangzhou
Period28/10/1229/10/12

Keywords

  • CENTRIST
  • Holistic approach
  • Object-based approach
  • Scene classification
  • Spatial pyramid matching (SPM)

ASJC Scopus subject areas

  • Artificial Intelligence
  • Computer Graphics and Computer-Aided Design

Cite this