KOHONEN NETWORKS FOR MULTISCALE IMAGE SEGMENTATION

An approach is developed to multiscale image segmentation, based on pi xel classification by means of a Kohonen network. An image is describe d by assigning a feature pattern to each pixel, consisting of a scaled family of differential geometrical invariant features. The invariant feature pattern representation of a training image is input to a Kohon en network to obtain a description of the feature space in terms of so -called prototypical feature patterns (the weight vectors of the netwo rk). Supervised labelling of these prototypical feature patterns may b e accomplished using classes derived from an a priori segmentation of the training image. We can segment any image similar to the training i mage by comparing the feature pattern representation of each pixel wit h all weight vectors, and assigning each pixel the class of the best m atching weight vector. In our study, we evaluated the benefit of apply ing features at multiple scales, as well as the effects of first- and second-order information on the results.