Title: A regularized model-based clustering method for image classification
Authors: Ying Zhu - National Institute of Education, Nanyang Technological University (Singapore) [presenting]
Abstract: Finite mixture models provide a flexible probabilistic modeling tool to handle heterogeneous data with a finite number of unobserved components. They are employed for model-based clustering in image classification. In high-dimensional spectroscopic data settings, spectral variable selection is both challenging and important to enable the feasibility of multivariate distribution fitting, especially for use in a real-time image classification. A regularized model-based clustering model is presented which enables an automatic selection of a small number of informative spectral variables for image classification. This model is on real life spectroscopic image data. The well-performed selection of spectral features leads to improve the classification accuracy, as well as to substantial reduce the clustering model complexity, and to provide better image representation.