Abstract:

Parametric mixture models for clustering algorithm depend too much on the prior assumptions and the orthogonal series density estimator is not a probability density function. To overcome these problems, a new image clustering algorithm based on normalized B-spline density model is proposed. A non-parametric mixture models based on normalized B-spline density function is designed, and NNBEM (Non-parametric B-spline Expectation Maximum) algorithm is used to
 estimate the unknown parameter of the density model, and the image clustering is in accordance with the Bayesian criterion. This algorithm dose not require any prior assumptions on the model, and it can effectively overcome the problem of the inconsistency between the parametric mixture models and the actual distribution. Some experiments about artificial data and real images are tested. These results show that the clustering method based on normalized B-spline density model is better than other algorithms.

Key words: computer image processing, clustering algorithm, B-spline density function, mixture model, Bayesian criterions