Feature Detection with Automatic Scale Selection

  • Authors:

  • Tony Lindeberg

  • Affiliations:

  • Computational Vision and Active Perception Laboratory (CVAP), Department of Numerical Analysis and Computing Science, KTH (Royal Institute of Technology), S-100 44 Stockholm, Sweden. E-mail: ton ...

  • Venue:
  • International Journal of Computer Vision
  • Year:
  • 1998

Quantified Score

Hi-index 0.01

Visualization

Abstract

The fact that objects in the world appear in different waysdepending on the scale of observation has important implications ifone aims at describing them. It shows that the notion of scaleis of utmost importance when processing unknown measurement data byautomatic methods. In their seminal works, Witkin (1983) andKoenderink (1984) proposed to approach this problem by representing image structures at different scales in a so-calledscale-space representation. Traditional scale-space theory buildingon this work, however, does not address the problem of how to select local appropriate scales for further analysis. This articleproposes a systematic methodology for dealing with this problem. Aframework is presented for generating hypotheses aboutinteresting scale levels in image data, based on a general principlestating that local extrema over scales of differentcombinations of γ-normalized derivatives are likelycandidates to correspond to interesting structures. Specifically, itis shown how this idea can be used as a major mechanism in algorithmsfor automatic scale selection, which adapt the local scales ofprocessing to the local image structure.Support for the proposed approach is given in terms ofa general theoretical investigationof the behaviour of the scale selection method under rescalings of the input pattern and by integrationwith different types of early visual modules,including experiments on real-world and synthetic data.Support is also given by a detailed analysis of howdifferent types of feature detectors perform when integrated with a scale selection mechanism and then applied to characteristic model patterns.Specifically, it is described in detail how the proposed methodologyapplies to the problems of blob detection, junction detection,edge detection, ridge detection and local frequency estimation.In many computer vision applications, the poor performance ofthe low-level vision modules constitutes a major bottleneck.It is argued that the inclusion of mechanisms for automatic scaleselection is essential if we are to construct vision systems toautomatically analyse complex unknown environments.