Comparing the Performance of Connectionist and Statistical Classifiers on an Image Segmentation Problem