A new unsupervised image segmentation algorithm based on deterministic annealing EM

• Published in: IEEE International Conference on Robotics, Intelligent Systems and Signal Processing, 2003. Proceedings. 2003 Vol. 1; pp. 600 - 604 vol.1
• Main Authors: Jiaqiang Zhong, Runsheng Wang
• Format: Conference Proceeding
• Language: English
• Published: IEEE 2003
• Subjects: Annealing; Convergence; Image processing; Image segmentation; Maximum likelihood estimation; Parameter estimation; Partitioning algorithms; Robots; Signal processing algorithms; Thermodynamics
• ISBN: 078037925X; 9780780379251
• DOI: 10.1109/RISSP.2003.1285642

A new unsupervised image segmentation algorithm based on deterministic annealing EM (DAEM) is proposed in this paper. The method is based on maximum likelihood (ML) estimation. Image is considered as a mixture of multi-variant normal densities and the number of densities is assumed to know. In order to obtain the parameters of densities, deterministic annealing EM algorithm is introduced. In DAEM algorithm, the EM process is reformulated as the problem of minimizing the thermodynamic free energy by using a statistical mechanics analogy. Thus, The DAEM algorithm can overcome the local maximize problem of general EM algorithm. The proposed method is successfully applied to image segmentation experiments.