Parameters of classification

Further it is necessary to customize parameters of classification.

Classifier generalization level defines how much classified regions can differ from a pattern. The more the level of generalization, the less similar to a pattern the regions will be classified. It is recommended at the first classification to set a minimum level of the classifier generalization.

Window size determines diameter of a round floating window for which the statistical characteristics are calculated.

Unlike one-class classification, the multiclass classifier preliminarily calculates for each pixel the probability of belonging to each class. The winner is the class with maximum probability. The more the window size, the greater the probability of correct classification inside of polygon objects. However on border between regions of different classes the low-contrast areas often are incorrectly classified, as belonging to a class with higher contrast. Therefore, the window size should be as small as possible, but sufficient for accurate classification.  

If in the classification window there will be too many shadow pixels, the results of classification will have low reliability. Therefore, before classification of a window the number of shadow pixels is checked. If their number will exceed allowable percentage of shadow pixels in the window, then the window is not classified. Optimal value of this parameter is 30%.

By pressing the button Choice of interpretive features you call the dialog of customizing the interpretive statistical and textural features.