Elsevier

Signal Processing

Volume 57, Issue 1, February 1997, Pages 19-33
Signal Processing

Paper
Designing texture filters with genetic algorithms: An application to medical images

https://doi.org/10.1016/S0165-1684(96)00183-1Get rights and content

Abstract

The problem of texture recognition is addressed by studying appropriate descriptors in the spatial frequency domain. During a training phase a filter is configured to determine different classes of texture by the response of its correlation with the Fourier spectrum of training-image templates. This is achieved by genetic algorithm-based optimisation. The technique is tested on standard texture patterns and then applied to magnetic resonance images of the brain to segment the cerebellum from the surrounding white and grey matter. Comparisons with established texture recognition techniques are presented, which show that the proposed method performs as well as, or better than, traditional techniques for the chosen instances of standard and anatomical texture and has the advantage of not having to decide which texture measure to use for a specific image structure.

Zusammenfassung

In diesem Artikel wird das Problem der Texturerkennung dadurch gelöst, daβ geeignete Abbildungen im Raum-Frequenz-Bereich betrachtet werden. Verschiedene Texturklassen sollen mit Hilfe eines Filters erkannt werden, welches in einer Trainingsphase konfiguriert wird. Dabei wird die Korreliertheit der zu bestimmenden Klassen mit dem Fourierspektrum einiger Trainingsschablonen ausgenutzt. Die Optimierung des Filters basiert auf der Anwendung von Genetischen Algorithmen. Diese Technik wird zunächst an Standardtexturmustern getestet und dann auf Magnetresonanzbilder des Gehirns angewandt, um das Zerebellum von der umgebenden weiβen und grauen Materie zu segmentieren. Vergleiche mit anerkannten Verfahren zur Texturerkennung zeigen, daβ die vorgestellte Methode eine gleiche oder sogar bessere Leistungsfähigkeit als die traditionellen Techniken aufweist. Desweiteren besitzt sie den Vorteil der Unabhängigkeit der Meβmethode von der speziellen Struktur des zu untersuchenden Bildes.

Résumé

Le problème de reconnaissance de texture est abordé en étudiant des descripteurs appropriés dans le domaine fréquentiel spatial. Un filtre est configuré durant une phase d'apprentissage, qui permet de déterminer les différentes classes de textures à l'aide de la réponse de leur corrélation avec le spectre de Fourier de modèles d'images d'apprentissage. Ceci est effectué par optimisation basée sur un algorithme génétique. La technique est testée sur des motifs de textures standards, puis appliquée à des images par résonnance magnétique du cerveau, afin de segmenter le cerebellum de la matière grise et blanche qui l'entoure. Des comparaisons sont faites avec des techniques établies de reconnaissance de structures, qui montrent que la méthode proposée se comporte aussi bien, voire mieux, que les techniques traditionelles pour les exemples choisis de textures standard et anatomiques, et qu'elle a l'avantage de ne pas nécessiter de décision quant à la mesure de texture à utiliser pour une structure d'image spécifique.

References (47)

  • K.K. Benke et al.

    Convolution operators as a basis for objective correlates for texture perception

    IEEE Trans. System Man. Cybernet.

    (1988)
  • P. Brodatz
  • T.M. Caelli

    An adaptive computational model for texture segmentation

    IEEE Trans. System Man. Cybernet.

    (1988)
  • D. Calloway

    Using a GA to design binary phase only filter for pattern recognition

  • P.C. Chen et al.

    Segmentation by texture using a co-occurence matrix

    Comput. Graph. Image Process.

    (1979)
  • I. Daubechies

    Orthonormal bases of compactly supported wavelets

    Comm. Pure Appl. Math.

    (1988)
  • K. Delibasis

    Genetic algorithms for medical image analysis

  • K. Delibasis et al.

    Genetic algorithm implementation of stack filter design for image restoration

  • K. Delibassis, P.E. Undrill and G.G. Cameron, “Genetic algorithms applied to Fourier descriptor based geometric models...
  • D. Goldberg
  • R. Gonzalez et al.
  • R. Gupta et al.

    The use of texture analysis to delineate suspicious masses in mammography

    Physics in Medicine and Biology

    (1995)
  • R.M. Haralick

    Statistical and structural approaches to texture

  • Cited by (15)

    • Multivariate image analysis in the process industries: A review

      2012, Chemometrics and Intelligent Laboratory Systems
      Citation Excerpt :

      For example, the use of a linear scale wavelet transform produced similar distance between different froth status both in feature space and image space [65], and the Mexican hat wavelet function produced more repeatable results in monitoring changes of medication tablets [76]. Wavelet coefficients can be thought as correlation coefficients between input signals and a mother wavelet and thus judicious design of wavelet filters can produce remarkable performance enhancement in texture analysis [69,77]. The different tree structures of wavelet transform can also affect performance of an analysis.

    • Functional template-based SAR image segmentation

      2004, Pattern Recognition
      Citation Excerpt :

      Because the nature of genetic optimization is randomized search from a number of search points (the individuals), the approach is ideal for parallel implementation. GAs have been applied to optimize overall segmentation quality [1,5,12,22,23], segmentation parameter selection [4,5,24–27] and to feature design for pixel-level classification [28–30]. GAs have also been used to determine the optimal subset of features for pixel-level classification.

    • Evolutionary computation in medicine: An overview

      2000, Artificial Intelligence in Medicine
    View all citing articles on Scopus
    View full text