Omni Script Writer Identification Problem.

Date of Submission

December 2010

Date of Award

Winter 12-12-2011

Institute Name (Publisher)

Indian Statistical Institute

Document Type

Master's Dissertation

Degree Name

Master of Technology

Subject Name

Computer Science


Computer Vision and Pattern Recognition Unit (CVPR-Kolkata)


Garain, Utpal (CVPR-Kolkata; ISI)

Abstract (Summary of the Work)

OMNI script writer identification problem is basically a writer identification problem in which images of hand writing of different authors are there in our database, at this point we are given a sample image of an unknown writer and we have to tell who is the writer of this image sample.Normally in this kind of problem our database consists of single script writing for example Hindi, English, Bengali etc. And unknown sample also comes under the same script but in case of OMNI script writer identification problem we release this restriction and allow multi-script to be there in our database.Though it seems to be a generalization of single script writer identification problem to multi-script environment but this is not really the case and you yourself will start believing this as we will go on farther and farther into the detail of our discussion.As far as single script writer identification problem is concerned several excellent works have already been done for script like English, Bengali, Hindi, French etc.For most of the solution, they have used a common way of attacking this problem which is to use pattern recognition problem i.e. extraction of a set of features from the hand writing of known writer and then based on these features classify the writer of an unknown sample as one of the known writer.They have mainly concentrated on allograph level features on a script under consideration which have been extracted by segmenting the text into lines, words, characters, graphemes etc. The use of allograph level feature requires knowledge in a particular script i.e. how to segment word into character or graphemes etc. And therefore extension of the method based on allograph level features is not straight forward to tackle multi-script problem where writer may write in different scripts.So we need a completely different treatment to solve our problem and need to extract those kind of features which are script independent that’s why we need to go through the very low level at the pixel level of the image sample.Previous WorkUnlike single script writer identification problem not much work have been done in this area so far. An extremely good work for solving this problem was done by my supervisor itself Dr. Utpal Garain and Mr. Thierry Paquet.The Basic idea which these two persons have used is that they have viewed hand writing image as a sequence of pixel value and they have tried to predict value of a pixel location by using previous say n terms.Let for the kth location we want to predict its pixel value say yk by using previous n terms. Let yk can be written asYk = a1yk-1+a2yk-2+..............+anyk-n+€nLet these ai’s, i=1,2,.....,n are such that they are best to predict pixel value of all location then this coefficient vector (a1,a2,......,an) are said to be AR coefficient of this image or image signal. AR coefficient for each image represents that particular writer.


ProQuest Collection ID:

Control Number


Creative Commons License

Creative Commons Attribution 4.0 International License
This work is licensed under a Creative Commons Attribution 4.0 International License.


This document is currently not available here.