Harish Srinivasan - Research and Publications

My research interests are in the application of the following models.

Conditional Random Fields (CRF) - For image segmentation and labeling CRF.
Infinite Multi-variate Gaussian Mixture Models (IGMM) - For clustering high dimensional data
Gaussian Processes - For Bayesian regression (My latest fancy)
Latent Dirichlet Allocation (LDA) and Topic Models (Related to the domain where I am am working)

I will try to post some presentations, papers as I make progress in these

My Publications. These are sorted year wise. All of these publications are applications of machine learning in the areas of hand writing and finger-prints.
Conditional Random Fields. Here you can find links to papers, presentations on CRFs as well as my research work on using CRF's to automatically segment and label scanned documents.
Infinite Multi-variate Gaussian Mixture Models. I used Infinite multi-varite Gaussian Mixture Models to cluster signature images. Since images are of very high dimensions and the number of clusters for a given corpus in unknown, the scenario is ideal for modeling the data with mixture of Gaussians with dirichlet prior to the mixing proportions with infinite number of mixture components. A demo for the same on a toy data set in 3 dimension can be downloaded here link
Gaussian ProcessesThis is one of my latest fancies. I am amazed by the power of Gaussian processes in Bayesian analysis. Here is a link to a demo of mine replicating Carl Rasmussen's tutorial. link
Fingerprint verificationThis project involves developing new algorithms for fingerprint verification, and study of individuality models for the use of fingerprints as biometrics.
Handwriting analysisThis project involves the development of a software package named CEDAR-FOX that performs writer verification based on handwriting and provides for plenty of other tools for handwriting analysis and searching.
Other projects Some of the other projects that I was involved are in Arabic document retrieval using word image as queries and Automatic Essay Scoring.

Given two fingerprints, one called the input and other template or known, the fingerprint verification problem is to decide if input fingerprint is Genuine (belonging to the same finger as the templtae) or Impostor. There are vast amounts of literature available on fingerprint analysis as it has been around for more than a century now. Hence listing out the papers in this field is virtually an impossible task. You can take a look at my publications in this domain, by searching for the keyword "Fingerprint", in my publications section. Some of the sub-topics and the students with whoom I work are mentioned below.

Matching algorithms: Graduate student Gang Fang and myself are looking into improving the existing matching algorithms for fingerprints. A matching algorithm is one that outputs a score for any pair of fingerprints compared.
Parametric modelling : Methods for using the score, the matching algorithm produces, to make a decision of Genuine/Imposter fingerprint can be done in a number of ways. Particulary we are looking into using parametric methods that involve modelling the scores using Gamma distributions and the use of Likelihhod methods.
Partial fingerprints: Graduate student Prasad Phatak is addressing issue of how partial fingerprints affect matching algorithms and how these can be tuned to improve verification.
Individuality models: Graduate student Vinu Chandar and myself are studying the various existing individuality models on fingerprint verification. Individuality models strengthen or weaken the use of a particular bio-metric, and different models give different probabilities of two fingerprints being the same.
Fingerprint classification. We will in the future plan to look into the aspect of classifying a fingerprint as Whorl, Arch, Loop, etc., through the use of Conditional Random Fields.

Back to start

Handwriting analysis

This project is one of the biggest projects that I am working in, along with a big team of other graduate students. All the various sub parts to this project, involve in some or the other way, the use of statistical pattern recognition and machine learning to a large extent. Few topics in machine learning that are used in here are, Hidden Markov Models, Various parametric and non-parametric classifiers such as Naive Bayes, k-nearest neighbor; neural networks, several clustering methods and plenty of image processing techniques. Handwriting recognition/analysis has always been one of the most challenging tasks for pattern recognition since there is a lot of variability. The research place I am working currently CEDAR that stands for Center of Excellence for Document Analysis and Recognition, founded by Dr.Srihari is where a lot of ground work on handwriting analysis was made. Today 95% of the handwritten postal mails that are received in the U.S. are machine sorted automatically, by reading the handwritten zip code in the mails. This project was developed at CEDAR. After this, a new project was started called CEDAR-FOX (CEDAR FOrensic eXamination). This project involved developing a software tool, that is capable of many tasks, few of wihch are listed below. My publications in these areas can be referred by searching for key words such as "handwritten" ,"words","spotting","verification", "retrieval" and "machine learning" in my publications section.

Writer verification: Using ones handwriting to tell if two documents were written by the same or different person is the writer verification problem. The software currently uses a simple Naive Bayes classifier to combine a lot of features to attack this problem. On going research involves adding new discriminatory features to the existing system, and the use of better statistical models to achieve the classification.
Word spotting: A Content-Based Image Retrieval (CBIR) wherein a handwritten word image is used as a query, to retrieve other scanned documents from a repository, that contain the same word is the task of this project.
Signature verification/retrieval: Off-line signature verification to verify, if a signature is Genuine or Forgery is always of importance. Computing strong discriminatory features, and developing efficient machine learning algorithms to learn from a set of known Genuine signatures of a person, to later tell, if a new signature belongs to this person or not, is the focus of this research project.

Back to start

Other projects

A couple of other projects that I am a part of are listed below and my publications in these fields can be obtained by searching for keywords "arabic", "essay scoring" in my publications section.

Automatic Essay Scoring: This is a grand challenge in the field of pattern recognition. Having a machine to automatically grade handwritten answers is the objective of this research. Current research in this field is the area of Latent Semantic Analysis (LSA) and using effective Information Retrieval ideas such as Named Entity Tagging (NET), probabilistic and vector space models. Such techniques are already used to a large extent for text analysis. The process of recognizing the words of a handwritten answer (typically we are focussing on answers on reading comphrehension questions given to school kids), involves use of HMM's and Viterbi algorithm.
Arabic projects: This project involves retieval of scanned Arabic documents based on Arabic word query and recognition of Arabic words.

Back to start

My publications

2007

H. Srinivasan, S. Kabra, C. Huang and S. N. Srihari, "On Computing the Strength of Evidence for Writer Verification," Proc. International Conference on Document Analysis and Recognition (ICDAR-2007), Curitiba, Brazil, September 2007. pdf (not yet available)

S. Shetty, H. Srinivasan, and S. N. Srihari, "Handwritten Word Recognition using Conditional Random Fields," Proc. International Conference on Document Analysis and Recognition (ICDAR-2007), Curitiba, Brazil, September 2007 pdf (not yet available)

A. Bharadwaj, A. Singh, H. Srinivasan, and S. N. Srihari, "On the use of Lexeme features for writer verification," Proc. International Conference on Document Analysis and Recognition (ICDAR-2007), Curitiba, Brazil, September 2007. pdf (not yet available)

S. N. Srihari, C. Huang, H. Srinivasan and V. Shah, "On the Discriminability of the Handwriting of Twins," Submitted. pdf (not yet available)

S. N. Srihari, J. Collins, R. K. Srihari, H. Srinivasan and S. Shetty, "Automatic Scoring of Short Handwritten Essays in Reading Comprehension Tests, " Submitted. pdf (not yet available)

G. Fang, S. N. Srihari, H. Srinivasan and P. Phatak, "Use of Ridge Points in Partial Fingerprint Identification," in Biometric Technology for Human Identification IV: Proc. of SPIE, Orlando, FL, April 2007, Vol 6539, pp. 65390D-1 to 65390D-9. pdf (not yet available)

M. Arivazhagan, H. Srinivasan, and S. N. Srihari, "A Statistical Approach to Handwritten Line Segmentation", in . Document Recognition and Retrieval XIV, Proceedings of SPIE, San Jose, CA, February 2007 pdf

S. Shetty, H. Srinivasan, S. N. Srihari and M. Beal, "Use of Conditional Random Fields in Document Image Retrieval," in Proc. Document Recognition and Retrieval IV, Proceedings of SPIE, San Jose, CA, February 2007. pdf (not yet available)

S. N. Srihari, H. Srinivasan and M. Beal, "Machine Learning for Signature Verification," in Learning in Document Analysis and Recognition S. Marinai and H. Fujisawa (eds.), Springer, 2007. pdf (not yet available)

S. N. Srihari, R. K. Srihari, H. Srinivasan and P. Babu, "On the Automatic Scoring of Handwritten Essays," in Proceedings of International Joint Conference on Artificial Intelligence(IJCAI), Hyderabad, India, January 2007. pdf (not yet available)

S. N. Srihari, C. Huang, H. Srinivasan and V. A. Shah, "Biometric and Forensic Aspects of Digital Document Processing" Digital Document Processing, B. B. Chaudhuri (ed.), Springer, 2007.

pdf

S. N. Srihari and H. Srinivasan, "A Statistical Model for Biometric Verification," in Modeling and Simulation in Biometric Technology, S. N. Yanushkevich, et. al. (eds.), World Scientific Press, 2007 pdf(not yet available)

2006

S.Shetty, H.Srinivasan and S.N.Srihari "Use of Conditional Random Fields for Signature based retrieval of scanned documents in the proceedings of Descartes Conference on Mathematical Methods in Counterterrorism (2006)" pdf

G. Ball, S. N. Srihari and H. Srinivasan, "Segmentation-Free and Segmentation-Dependent Approaches to Arabic Word Spotting,"Proc. International Workshop on Frontiers in Handwriting Recognition (IWFHR-10), La Baule, France, October 2006 pdf (not yet available)

S. N. Srihari, H. Srinivasan, C. Huang and S. Shetty, "Spotting Words in Latin, Devanagari and Arabic Scripts," Vivek: Indian Journal of Artificial Intelligence. pdf.

S. N. Srihari, S. Shetty, S. Chen, H. Srinivasan, C. Huang, G. Agam, O. Frieder, "Document Image Retrieval using Signatures as Queries," Proc. Second IEEE International Conference on Document Image Analysis for Libraries, Lyon, France, April 27-28, 2006. pdf

H. Srinivasan, S. N. Srihari, M. Beal, G.Fang and P.Phatak "Comparison of ROC-based and Likelihood methods for Fingerprint Verification. " Proc. Biometric Technology for Human Identification III: SPIE Defense and Security Symposium, Orlando, FL, April 17-22, 2006. pdf.

S. N. Srihari, J. Collins, R. K. Srihari, P. Babu and H. Srinivasan "Automatic Scoring of Handwritten Essays using Latent Semantic Analysis" Proc. Document Analysis Systems Springer, Nelson, New Zealand, February 2006, pp. 71-83. pdf.

S. N. Srihari, H. Srinivasan, P. Babu and C. Bhole, "Spotting Words in Handwritten Arabic Documents," Proc. Document Recognition and Retrieval XIII(SPIE), San Jose, CA, January 2006, pp. 606702-1 to 606702-12. pdf.

2005

S. N. Srihari, H. Srinivasan, P. Babu and C. Bhole, "Handwritten Arabic Word Spotting using the CEDARABIC Document Analysis System," Proc. Symposium on Document Image Understanding (SDIUT 05), College Park, MD, November 2005. pdf.
H. Srinivasan, S. N. Srihari and M. Beal, "Signature Verification using Kolmogorov-Smirnov Statistic," Proc. International Graphonomics Society Conference (IGS), June 2005, Salerno, Italy, pp. 152-156.pdf.
H. Srinivasan, M. Beal and S. N. Srihari, "Machine Learning for Person Identification and Verification", SPIE Conference on Homeland Security, Orlando, FL, March 2005, Society of Photo Instrumentation Engineers (SPIE), pp. 574-586. pdf.
S. N. Srihari, C. Huang and H. Srinivasan, "A Search Engine for Handwritten Documents", Document Recognition and Retrieval XII, San Jose, CA, January 2005, Society of Photo Instrumentation Engineers (SPIE), pp. 66-75. pdf.

2004

S. N. Srihari, C. Huang and H. Srinivasan, "Content-Based Retrieval of Handwritten Document Images," Knowledge Based Computer Systems (KBCS 2004), Hyderabad, India, December 2004. pdf.

Back to start