Strategies for unsupervised multimedia processing

A computational technology based on artificial neural network architectures having a basis in self organization is discussed. Dynamically growing models of self-organization address the problem of inferring an appropriate number of inferring an appropriate of classes from the underlying data distribution and thus,exhibits greater flexibility than static approaches. Self-organizing tree map (SOTM) is an efficient approach in formulating a topological set of classes that attempt to maximize pattern discrimination while hierarchically parsing an underlying distribution. Directed SOTM and self-organizing hierarchical variance map (SOVHM) have been applied successfully to content-based image retrieval, and to the segmentation of 3-D microbial image data.