Morphological systems for character image processing and recognition

Min/max signal operations, common in morphological image analysis were applied to both feature extraction and classification of character images. A system is proposed that computes an improved version of the morphological shape-size histogram. It reduces sensitivity to stroke thickness, size, and rotation. For pattern classification, the class of min-max classifier, which generalizes Boolean DNF functions for real-valued inputs, is introduced. A least mean square (LMS) algorithm was used for practical training of min-max classifiers. Experimental results show that min-max classifiers were able to achieve error rates comparable with those of neural networks trained using backpropagation. The main advantages of the min-max/LMS algorithm are its simplicity and faster speed of convergence.<>