Text and script independent writer identification

Abstract

If a set of writers know writing of more than one scripts/languages, identification of such writers is difficult and challenging problem of research. One method is to design a script independent writer identification algorithm to identify the writer of underlying handwritten document. Hence a text and script independent method is proposed for identification of writer for handwritten scripts/languages using correlation and homogeneity properties of Gray Level Co-occurrence Matrices of the handwritten document images. The feature vector of size 40 is obtained from each input handwritten document image. Handwritten documents are collected from the same 100 writers in Roman, Kannada and Devanagari scripts. Using nearest neighbor classifier with modified 4-fold cross validation the results for writer identification are obtained. Identification accuracies are 82.75%, 82.75% and 85.25% when the handwritten documents are in only one script Roman, Kannada and Devanagari scripts respectively. The writer identification rates are 80.6250%, 83.75% and 84% respectively for Roman-Kannada, Roman-Devanagari and Kannada-Devanagari handwritten input documents. The writer identification rate is 82.1995% for the input documents of Roman-Kannada-Devanagari.