Before classification can happen, the system has to be trained using one or more sample images for each class. This is referred to as learning. New categories can also be added incrementally with their sample images without relearning the complete classification pattern.

Learning is performed by training the system. You can provide training samples in predefined folders on your hard disk containing an arbitrary number of sample documents in each folder. Each folder corresponds to one category or class of document. The Image Classifier uses these sample documents in the learning step to generate the patterns so unknown documents of the same type recognized later in the classification step. This means that you have to define the training samples first and then perform the learning before the you are ready for classification.