Arabic (Indian) Handwritten Digits Recognition Using Multi feature and KNN Classifier

Abstract

This paper presents an Arabic (Indian) handwritten digit recognition system based on combining multi feature extraction methods, such a upper_lower profile, Vertical _ Horizontal projection and Discrete Cosine Transform (DCT) with Standard Deviation σi called (DCT_SD) methods. These features are extracted from the image after dividing it by several blocks. KNN classifier used for classification purpose. This work is tested with the ADBase standard database (Arabic numerals), which consist of 70,000 digits were 700 different writers write it. In proposing system used 60000 digits, images for training phase and 10000 digits, images in testing phase. This work achieved 97.32% recognition Accuracy.