Integrating Signal and Language Context to Improve Handwritten Phrase Recognition: Alternative Approaches

Djamel Bouchaffra, Eugene Koontz, V. Krpasundar, Rohini K. Srihari, Sargur N. Srihari
Proceedings of the Sixth International Workshop on Artificial Intelligence and Statistics, PMLR R1:47-54, 1997.

Abstract

Handwritten phrase recognition is an important and difficult task. Recent research in this area has fo- cussed on utilising language context to improve recognition performance, without taking the information from the input signal itself into proper account. In this paper, we adopt a Bayesian approach to solving this problem. The Bayesian framework allows us to integrate signal-level information from the actual input with the linguistic context usually used in post-processing the recogniser’s output. We demonstrate the validity of a statistical approach to integrating these two sources of information. We also analyse the need for improvement in performance through innovative estimation of informative priors, and describe our method for obtaining agreement from multiple experts for this task. We compare the performance of our integrated signal-language model against existing "language-only" models.

Cite this Paper

BibTeX
@InProceedings{pmlr-vR1-bouchaffra97a, title = {Integrating Signal and Language Context to Improve Handwritten Phrase Recognition: Alternative Approaches}, author = {Bouchaffra, Djamel and Koontz, Eugene and Krpasundar, V. and Srihari, Rohini K. and Srihari, Sargur N.}, booktitle = {Proceedings of the Sixth International Workshop on Artificial Intelligence and Statistics}, pages = {47--54}, year = {1997}, editor = {Madigan, David and Smyth, Padhraic}, volume = {R1}, series = {Proceedings of Machine Learning Research}, month = {04--07 Jan}, publisher = {PMLR}, pdf = {http://proceedings.mlr.press/r1/bouchaffra97a/bouchaffra97a.pdf}, url = {https://proceedings.mlr.press/r1/bouchaffra97a.html}, abstract = {Handwritten phrase recognition is an important and difficult task. Recent research in this area has fo- cussed on utilising language context to improve recognition performance, without taking the information from the input signal itself into proper account. In this paper, we adopt a Bayesian approach to solving this problem. The Bayesian framework allows us to integrate signal-level information from the actual input with the linguistic context usually used in post-processing the recogniser’s output. We demonstrate the validity of a statistical approach to integrating these two sources of information. We also analyse the need for improvement in performance through innovative estimation of informative priors, and describe our method for obtaining agreement from multiple experts for this task. We compare the performance of our integrated signal-language model against existing "language-only" models.}, note = {Reissued by PMLR on 30 March 2021.} }
Endnote
%0 Conference Paper %T Integrating Signal and Language Context to Improve Handwritten Phrase Recognition: Alternative Approaches %A Djamel Bouchaffra %A Eugene Koontz %A V. Krpasundar %A Rohini K. Srihari %A Sargur N. Srihari %B Proceedings of the Sixth International Workshop on Artificial Intelligence and Statistics %C Proceedings of Machine Learning Research %D 1997 %E David Madigan %E Padhraic Smyth %F pmlr-vR1-bouchaffra97a %I PMLR %P 47--54 %U https://proceedings.mlr.press/r1/bouchaffra97a.html %V R1 %X Handwritten phrase recognition is an important and difficult task. Recent research in this area has fo- cussed on utilising language context to improve recognition performance, without taking the information from the input signal itself into proper account. In this paper, we adopt a Bayesian approach to solving this problem. The Bayesian framework allows us to integrate signal-level information from the actual input with the linguistic context usually used in post-processing the recogniser’s output. We demonstrate the validity of a statistical approach to integrating these two sources of information. We also analyse the need for improvement in performance through innovative estimation of informative priors, and describe our method for obtaining agreement from multiple experts for this task. We compare the performance of our integrated signal-language model against existing "language-only" models. %Z Reissued by PMLR on 30 March 2021.
APA
Bouchaffra, D., Koontz, E., Krpasundar, V., Srihari, R.K. & Srihari, S.N.. (1997). Integrating Signal and Language Context to Improve Handwritten Phrase Recognition: Alternative Approaches. Proceedings of the Sixth International Workshop on Artificial Intelligence and Statistics, in Proceedings of Machine Learning Research R1:47-54 Available from https://proceedings.mlr.press/r1/bouchaffra97a.html. Reissued by PMLR on 30 March 2021.

Related Material