Tsallis entropy is a q parameter extension of Shannon entropy. By maximizing Tsallis entropy within the framework of FCM, a membership function similar to the statistical mechanical distribution functions is obtained. The membership function can be combined with the deterministic annealing method.

One of the major issue of the Tsallis entropy maximization method is that how to determine an appropriate q value according to a data distribution is not clear.

In this article, we propose a method in which q is increased while lowering the annealing temperature so as to minimize the objective function, because q strongly affects the extent of the membership function.

Numerical experiments are performed and the obtained results indicate that the proposed method works properly and the q value can be adjusted so as to make the membership function fit to the data distribution.

fuzzy c-means, Tsallis entropy maximization, deterministic annealing.