A novel autonomous feature clustering model for image recognition

In order to realize human-like image recognition system, we introduce an architecture with separation of extracting and clustering features from detecting features. Me also propose a novel autonomous clustering model that attaches an adaptive cluster determination algorithm, which enables superior cluster determination even for higher dimension vectors like real world images, on the Kohonen's Self-Organizing feature Map (SOM). By this algorithm, SOM weight vectors are converted to extremely lower dimensional vectors, which just consist of meaningful components to describe clusters. Therefore, we can execute autonomous determination of cluster boundaries easily. As a result, our proposed clustering model shows better performance than conventional techniques. Furthermore, feature detectors in our architecture is self-organized by the clustered sets of features which is autonomously clustered in our model.