Model-based recognition of multiple deformable objects using a game-theoretic framework


BOZMA AYDIN H. I., Duncan J. S.

12th International Conference on Information Processing in Medical Imaging, IPMI 1991, Wye, İngiltere, 7 - 12 Temmuz 1991, cilt.511 LNCS, ss.358-372, (Tam Metin Bildiri)

  • Yayın Türü: Bildiri / Tam Metin Bildiri
  • Cilt numarası: 511 LNCS
  • Doi Numarası: 10.1007/bfb0033765
  • Basıldığı Şehir: Wye
  • Basıldığı Ülke: İngiltere
  • Sayfa Sayıları: ss.358-372
  • Anahtar Kelimeler: Image processing systems, Nash equilibrium, Parallel algorithms and architectures, Sensor fusion
  • Boğaziçi Üniversitesi Adresli: Evet

Özet

Image analysis systems aimed at robust segmentation via model-based recognition of multiple deformable objects require integration of a variety of modules. One approach to modularity and integration is based on decomposing the task into a set of coupled modules with their corresponding objectives, and then integrating them within a unifying framework. Although both theoretical and computational considerations indicate the importance of adhering to the multiple coexisting nature of the objectives, previous approaches to integration have been limited in this aspect. In (Bozma, 1990), an integration framework which overcomes this problem, is developed based on game-theoretic concepts. The contribution of this paper is to describe a segmentation system aimed at model-based recognition of multiple deformable objects, in which the modules are integrated using the game-theoretic approach. A secondary contribution is that the model used in object recognition, which is defined as a set of independent, yet coupled deformable objects, constitutes an extension of the previous models. Our experiments with several biomedical images using anatomical atlases as the basis for these models demonstrate the power of this system in medical applications.