In this paper we present an unsupervised color image segmentation algorithm based on statistical models. We have adopted the Gaussian mixture model to represent the distribution of color feature vectors. A novel deterministic annealing EM and mean field theory from statistical mechanics are used to compute the posterior probability distribution of each pixel and estimate the parameters of the Gaussian Mixture Model. We describe the noncontexture segmentation algorithm that uses a deterministic annealing approach and the contexture segmentation algorithm that uses the mean field theory. The experimental results show that the deterministic annealing EM and mean field theory provide a global optimal solution for the maximum likelihood estimators and that these algorithms can efficiently segment the real image.

A standard of Advanced Storage Magneto Optical (AS-MO) having a 6 Gbyte capacity in a 120 mm-diameter single side disk was established by using a magnetically induced superresolution readout method. Transition from in-plane to perpendicular magnetization for exchange-coupled readout layer (GdFeCo) and in-plane magnetization mask layer (GdFe) of the AS-MO disk has been investigated using the noncontinuous model. The readout resolution was sensitive to the thickness of the readout layer. To evaluate readout characteristics of AS-MO disks, the simulation using micro magnetics model was performed and the readout layers were designed. The readout characteristics of the AS-MO disk is improved by making the readout layer thinner.

The mean field theory has been recognized as offering an efficient computational framework in solving discrete optimization problems by neural networks. This paper gives a formulation based on the information geometry to the mean field theory, and makes clear from the information-theoretic point of view the meaning of the mean field theory as a method of approximating a given probability distribution. The geometrical interpretation of the phase transition observed in the mean field annealing is shown on the basis of this formulation. The discussion of the standard mean field theory is extended to introduce a more general computational framework, which we call the generalized mean field theory.