Bayesian learning of generalized Gaussian mixture models on biomedical images

Abstract

In the context of biomedical image processing and bioinformatics, an important problem is the development of accurate models for image segmentation and DNA spot detection. In this paper we propose a highly efficient unsupervised Bayesian algorithm for biomedical image segmentation and spot detection of cDNA microarray images, based on generalized Gaussian mixture models. Our work is motivated by the fact that biomedical and cDNA microarray images both contain non-Gaussian characteristics, impossible to model using rigid distributions like the Gaussian. Generalized Gaussian mixture models are robust in the presence of noise and outliers and are more flexible to adapt the shape of data. © 2010 Springer-Verlag.