Image classification with structured self-organization map

Adaptive processing of structured data using a supervised learning scheme has been successfully applied to many domains, e.g., molecular biology, image classification and retrieval. A self organizing map (SOM) type algorithm for processing of structured data using an unsupervised learning approach has previously been proposed. We present an approach using quadtree representation to extract an image structure, and the application of such a structured SOM to image classification problems. Encouraging results achieved by using only six simple visual features show that the structured SOM works well for structural information