Image segmentation applications with unsupervised neural networks

Abstract

Image segmentation is the most difficult task of low level computer vision applications. The performance of the segmentation algorithm affects the subsequent applications performances such as high level pattern recognition applications. In this study we examine 1-D and 2-D unsupervised artificial neural networks (Kohonen network) and we evaluate the performances of the examined structures over image segmentation applications. Experimental studies show the advantages and disadvantages of the Kohonen networks. © 2005 IEEE.