Writer Identification based on Writing Individuality and Combination of Features

Abstract

Individuality in handwriting occurs due to genetic as well as memetic factors. Memetic or cultural factors originate from training-related influences on pen-grip style and writing-strokes. This work presents a writer identification scheme using individuality of handwriting. Performance of such a scheme is greatly influenced by judicious selection off eatures and their combinations. In this work three types of features are selected, namely, line based, word based and character based. Importance of these features in writer identification is explored in the work. A dataset of English handwritten text samples written by subjects of Bengali origin and Bengali-medium schooling background is used for experimentation. Results are given for different combinations of features mentioned above. Emphasis is given on stroke features which include shapes like doughnut, hump, stick and stem-loop. They are found commonly in English lowercase letters. Best results obtained by using classifiers MLP and K-STAR are 93.54 percent and 95.69 percent respectively.