Offline Writer Identification from Isolated Characters Using Textural Features

Study on behavioural biometric has gained renewed interest from researchers in recent years. Writer identification and verification is one of the areas that has promising prospect in real-life applications like forensic, security, access control, HOCR (Handwritten Optical Character Recognizer), etc. We could not find any complete system for writer identification/verification on Indic scripts including Bangla. In this proposed method, we have modified and evaluated the performance of FFT (Fast Fourier Transform), GLCM (Gray-Level Co-occurrence Matrix), DCT (Discrete Cosine Transform) on our general unconstrained Bangla character database. The database is a collection of total 53250 Bangla characters (38250 alphabets + 7500 Bangla numerals + 7500 Bangla vowel modifiers) from 150 writers with 5 sets from each writer. Modification on FFT, GLCM and DCT to use as textural features and combination of those features produces promising results. The results show that our method is comparable with other available works and capable of handling large volume of data.