Evolutionary Image Analysis and Signal Processing

Volume 213 of the series Studies in Computational Intelligence pp 21-40

Detecting Scale-Invariant Regions Using Evolved Image Operators

  • Leonardo TrujilloAffiliated withCentro de Investigación Científica y de Educación Superior de Ensenada
  • , Gustavo OlagueAffiliated withCentro de Investigación Científica y de Educación Superior de Ensenada

* Final gross prices may vary according to local VAT.

Get Access


This chapter describes scale-invariant region detectors that are based on image operators synthesized through Genetic Programming (GP). Interesting or salient regions on an image are of considerable usefulness within a broad range of vision problems, including, but not limited to, stereo vision, object detection and recognition, image registration and content-based image retrieval. A GP-based framework is described where candidate image operators are synthesized by employing a fitness measure that promotes the detection of stable and dispersed image features, both of which are highly desirable properties. After a significant number of experimental runs, a plateau of maxima was identified within the search space that contained operators that are similar, in structure and/or functionality, to basic LoG or DoG filters. Two such operators with the simplest structure were selected and embedded within a linear scale space, thereby making scale-invariant feature detection a straightforward task. The proposed scale-invariant detectors exhibit a high performance on standard tests when compared with state-of-the-art techniques. The experimental results exhibit the ability of GP to construct highly reusable code for a well known and hard task when an appropriate optimization problem is framed.