A three-level classifier: Fuzzy C Means, Support Vector Machine and unique pixels for Arabic handwritten digits

In this study, we present a classification approach for handwritten Arabic digits (symbols). Like numbers in other languages, Arabic numbers consists of nine digits. Character images of Arabic digits are similar in the sense that one single classifier will not give a reliable classification rate. Therefore, the implementation of more levels of classification is important for the realization. We introduce Fuzzy C-Means based classifier for the lower level and Support Vector Machine SVM for the second level when more details are required and finally confirmation of classification will be through unique pixels. The unique pixel method forms the third classification level. It works on determining the pixel areas that are unique to each digit. The unique pixel method decision is compared with decision of the lower classifier (FCM) and top classifier (SVM). The algorithm is tested on 3510 images. The overall testing accuracy reported is 88%.