Fast segmentation of large images

  • Authors:

  • David J. Crisp;Peter Perry;Nicholas J. Redding

  • Affiliations:

  • Intelligence, Surveillance & Reconnaissance Division, Defence Science & Technology Organisation, PO Box 1500, Edinburg, South Australia;SYDAC Pty. Ltd., 113-115 King William Street, Adelaide S.A.;Intelligence, Surveillance & Reconnaissance Division, Defence Science & Technology Organisation, PO Box 1500, Edinburg, South Australia

  • Venue:
  • ACSC '03 Proceedings of the 26th Australasian computer science conference - Volume 16
  • Year:
  • 2003

Quantified Score

Hi-index 0.00

Visualization

Abstract

The processing of large images is a generic problem in wide area surveillance. An important difficulty is that many image processing algorithms are global rather than local and hence can be infeasible due to the required computing time or memory resources when processing very large images. Consequently there is a need to break such image processing problems into smaller pieces. A similar need also arises when the requirement is to process the imagery on-line as it is being collected. Here, we consider the particular problem of image segmentation. The approach we take is to divide the large image into smaller overlapping tiles. We segment each tile separately and then patch the results together. The main contribution of this paper is our answer to the question of how to handle ambiguities in the overlapping areas. Our solution is applicable to any segmentation algorithm which is based on region merging. The particular algorithm we use is known as the Full Lambda Schedule Algorithm (FLSA). We also include a description of some refinements to the original algorithm which provide speed and efficiency improvements.