Farsi/Arabic Handwritten from Machine-Printed Words Discrimination

Separating handwritten texts from machine-printed materials is a desirable task towards a general document analysis system. In this paper, we proposed a simple and effective method to discriminate handwritten from machine-printed words in Farsi/Arabic documents. After finding word blocks, three different feature sets were extracted. They include two well-established features, previously used for Latin handwritten from machine-printed text separation, and a new feature, called baseline profile. Then, extracted features were combined together to obtain a feature vector with 34 elements. SVM and KNN classifiers were utilized to separate handwritten and machine-printed words. To evaluate the proposed method, some special forms, designed for word separation, were used. Experimental results show that our system differentiates between handwritten and machine-printed words with the overall accuracy of 97.1%.