Image segmentation by fuzzy rule and Kohonen-constraint satisfaction fuzzy c-mean

Abstract

In this paper, we present a new algorithm that can segment a fuzzy data. This method is based on fuzzy logic and neural network, and a concept of constraint satisfaction problem (CSP). Firstly, pre-processing by creating a new pixel value on fuzzy rule has ability to describe the effect of neighborhood pixel in degree of membership value or linguistic variables which are utilized to activate a rule base. Secondly, the feature extraction employs Kohonen feature mapping (SOM), which can learn a feature without prior knowledge. Finally, a fuzzy c-mean (FCM) adapts to CSP structure by constituting a interconnection weight for all membership value of fuzzy c-mean with a global consistency situation. The result of this method have been tested on a varieties images.