Towards Interpretation of Self Organizing Map For Image Segmentation

This paper provides an effective framework to interpret the data of self-organizing map (SOM). It is known that data clustering SOM is one of the most popular neural networks used for image segmentation. The interpretation of SOM output has to be further processed for obtaining segmented image. In the proposed method the SOM is used with extracted features data and the output is analyzed to obtain the best match units (BMU). The highest winners of BMU's are considered as a cluster representative. In the second stage the winner BMU's are filtered to derive the best cluster representative based on number of clusters and predefined Euclidean distance between the winners. Finally the clustering labeling is carried out with reference to cluster representative. This method has been tested with Berkeley's database and preliminary results are promising. The results have also been compared with FCM and K Means algorithms.