WRITER IDENTIFICATION FOR DEVANAGARI SCRIPT

Abstract

Writer identification is the task of formative the person whose handwritten sample is available in a set of writings, collected from multitude of writers. This has useful applications in many areas, conspicuously in forensic analysis. The task of writer identification is quite difficult due to marginal variations found in different handwritten samples from same person/writer. Several identification algorithms have been recommended so far which are mostly for non-Indic writings. Research into writer identification has been focused on two streams, off-line and on-line writer identification. Generally it is believed that text-independent writer identification is more thought-provoking than text-dependent writer identification. Text-independent Offline writer recognition is more stimulating than online writer recognition. Here we propose a system which extracts the simple writer specific features from the scanned handwritten documents by different writers and use them to recognize the writer. Based on the idea that has been presented in the previous studies, here we assume handwriting as texture image and a set of features which are based on multi-channel Gabor filters and GLCM (Gray Level Cooccurrence Matrix) are extracted from preprocessed image of documents. First order statistical features are also extracted from the documentation. Substantially, the property of proposed method is using of the bank of Gabor filters which is appropriate for structure of Devanagari handwritten texts and vision system. As there is no predefined dataset of Devanagari Handwritten reports, so for this we first made our own particular database. The database comprises of pictures of composing of 45 authors. Each author composes five same constituent in five pages, it makes an aggregate of 225 records and every archive contains 102 words. Fifty percent of the aggregate authors of the database are female and the remaining authors are male with the age gathering changing from 21 years to 60 years. Out of 225 pictures, 180 content pictures are utilized for training and rest is utilized for testing. We have evaluated features from the 2-D Gabor filter, GLCM and first order statistical method and the classifier used for identification of writer is k-nearest neighbor (k-NN). The above approach is tested for our database and experimental results are encouraging.