This research presents a comparison between two methods of learning-classification, the first one is a probabilistic and unsupervised which is the Hidden Model Markov (HMM), while the second is a statistic and supervised that is the Support Vectors Machine (SVM). These techniques are used for printed Eastern Arabic numerals recognition, in different situation: rotated and translated or resized and noisy. In the pre-processing phase we have used the thresholding technique while in the features extraction we exploited the Krawtchouk Invariant Moment (KIM). In fact, in order to make a precise comparison between these two methods, we have introduced two new concepts which are the threshold and the interval of stability of each numeral and for each of these two methods. The simulation results that we obtained demonstrates that SVM is more performing than the HMM technique in this recognition system.
Show abstract
Full Text
