A Supervised Classifier for Multispectral and Textured Images Based on an Automated Region Growing Algorithm

A couple of supervised classifiers to segment optical multispectral images and textured radar images has been developed. In both classifiers, an automated regiongrowing algorithm delineates the training sets. Optimum statistics for defined classes are derived from the training sets. This algorithm handles three parameters: an initial pixel seed, a window and a threshold for each class. A suitable pixel seed is manually implanted through visual inspection of the image classes. The optimum value for the window and the threshold are obtained from spectral or texture distances. These distances are calculated from mathematical models of spectral and textural separabilities. A pixel is incorporated into a region if a spectral or texture homogeneity criterion is satisfied in the pixel-centered window for a given threshold. In this scheme, a region grows as much as possible but maintains the overlap with other regions in a minimum. The homogeneity criterion is obtained from the models of spectral and textures distances. The set of pixels forming a region represents a statistically valid sample of a defined class signaled by the initial pixel seed. The grown regions constitute therefore optimum training sets for each class. The statistical behavior of these training sets is used to classify the pixels of the image in one member of a set of classes. Comparing the statistical behavior of a sliding window with that of each class does the classification. The size of this window is the same as the one employed in the region-growing algorithm. The centered pixel of the sliding window is labeled as belonging to a class if its spectral or texture distance is a minimum to the class. Such distance is evaluated using the statistical content of the class and the sliding window as input to the model of spectral or textural separability. A series of examples, employing synthetic and natural images, are presented to show the value of this classifier. The goodness of the segmentation is evaluated by means of the Kappa coefficient and a matrix of distances derived from the mentioned model.