Self-Organization of Feature Columns and its application to Object Classification

We propose a computational model for the self-organization of feature columns based on a modular framework. The proposed model consists of several modules; each module is composed of a collection of Gaussian units. The combination of hierarchical competition within and between modules and a smoothness constraint finds out continuity among input patterns, and topographically maps these series of patterns into different modules. Computer simulations show an example of self-organization and object classification by combining the created feature columns.