A multiscale region detector

Dorothea Blostein, Narendra Ahuja

Research output: Contribution to journalArticlepeer-review


A multiscale region detector is derived from properties of a {down triangle, open}2G (Laplacian-of-Gaussian) scalespace. The detector finds all image regions that have small gray-level variation relative to a neighborhood of their size. These regions may be nested. Since the region detector is designed as a coarse estimator of region size and shape, it is better suited for texture element extraction than for general image segmentation. Uniform gray level regions are represented by a configuration of best-fitting disks of appropriate sizes and locations contained within the regions. The sizes and locations of best disks are estimated from the observed response of the {down triangle, open}2G operator across the image and across scales, using an analytic expression for the response of {down triangle, open}2G to disks. Both positive-contrast and negative-contrast regions are detected, and an estimate of region contrast is computed. Results are shown for a variety of natural textures.

Original languageEnglish (US)
Pages (from-to)22-41
Number of pages20
JournalComputer Vision, Graphics and Image Processing
Issue number1
StatePublished - Jan 1989

ASJC Scopus subject areas

  • General Environmental Science
  • General Engineering
  • General Earth and Planetary Sciences


Dive into the research topics of 'A multiscale region detector'. Together they form a unique fingerprint.

Cite this