Multi-scale feature identification using evolution strategies

  • Authors:

  • Xiaojing Yuan;Jian Zhang;Xiaohui Yuan;Bill P. Buckles

  • Affiliations:

  • Mechanical Engineering Department, Tulane University, New Orleans, LA 70118, USA;Electrical Engineering and Computer Science Department, Tulane University, New Orleans, LA 70118, USA;Electrical Engineering and Computer Science Department, Tulane University, New Orleans, LA 70118, USA;Electrical Engineering and Computer Science Department, Tulane University, New Orleans, LA 70118, USA

  • Venue:
  • Image and Vision Computing
  • Year:
  • 2005

Quantified Score

Hi-index 0.00

Visualization

Abstract

An image usually contains a number of different features and regions. Many image-related applications, such as content-based image retrieval and MRI-based diagnosis, often require the ability to identify and mark features within the image. For images containing a specific sort of feature (e.g. convective storm) or region (e.g. earthquake debris), that feature or region is always located adjacent to other features and regions on the image. A generic framework for automatically identifying features in images based on evolutionary computation is proposed here. The significant characteristic of the method is that it does not require segmentation. We use evolution strategies as the optimization algorithm to identify features. The system is based on a conjecture that certain filters will give prominent responses to certain features. The identified features are represented as regions enclosed within the chosen search structure-the ellipse. By defining filter response criteria as the fitness function, evolution strategies succeeds in finding the feature in a much more efficient way than, say, segmentation.