Image recognition with occlusions

Tyng Luh Liu, Mike Donahue, Davi Geiger, Robert Hummel

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

Abstract

We study the problem of how to detect “interesting objects” appeared in a given image, I. Our approach is to treat it as a function approximation problem based on an over-redundant basis. Since the basis (a library of image templates) is over-redundant, there are infinitely many ways to decompose I. To select the “best” decomposition we first propose a global optimization procedure that considers a concave cost function derived from a “weighted Lpnorm” with 0 < p ≤ 1. This concave cost function selects as few coefficients as possible producing a sparse representation of the image and handle occlusions. However, it contains multiple local minima. We identify all local minima so that a global optimization is possible by visiting all of them. Secondly, because the number of local minima grows exponentially with the number of templates, we investigate a greedy “LpMatching Pursuit” strategy.

Original languageEnglish (US)
Title of host publicationComputer Vision – ECCV 1996 - 4th European Conference on Computer Vision, Proceedings
EditorsBernard Buxton, Roberto Cipolla
PublisherSpringer Verlag
Pages556-565
Number of pages10
ISBN (Print)3540611223, 9783540611226
DOIs
StatePublished - 1996
Event4th European Conference on Computer Vision, ECCV 1996 - Cambridge, United Kingdom
Duration: Apr 15 1996Apr 18 1996

Publication series

NameLecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)
Volume1064
ISSN (Print)0302-9743
ISSN (Electronic)1611-3349

Other

Other4th European Conference on Computer Vision, ECCV 1996
CountryUnited Kingdom
CityCambridge
Period4/15/964/18/96

ASJC Scopus subject areas

  • Theoretical Computer Science
  • Computer Science(all)

Fingerprint Dive into the research topics of 'Image recognition with occlusions'. Together they form a unique fingerprint.

Cite this