论文信息 - A Statistical Correction-Rejection Strategy for OCR Outputs in Persian Personal Information Forms

A Statistical Correction-Rejection Strategy for OCR Outputs in Persian Personal Information Forms

In this paper, a MAP statistical modeling * approach has been utilized to correct and verify Persian names and surname OCR outputs. In addition, an efficient Neural Network based rejection method has been presented and tested. Due to large variety of Persian surnames, a statistical grammar has been added to the MAP strategy, to make new surnames, which are not included in the dictionary. The model has been analytically formulated and practically implemented. The achieved results show a large character and word error reduction while the calculation increase is negligible in comparison with character recognition complexity.

Ramin Mehran | Farbod Razzazi | Amin Shali

[1] William A. Gale,et al. Good-Turing Smoothing Without Tears , 2001 .

[2] M. Degroot. Optimal Statistical Decisions , 1970 .

[3] A. Fatholahzadeh. Lexicon Base in Persian for Object ManipulationA , 1993 .

[4] Juan Carlos Pérez-Cortes,et al. Rejection strategies and confidence measures for a k-NN classifier in an OCR task , 2002, Object recognition supported by user interaction for service robots.

[5] C. K. Chow,et al. On optimum recognition error and reject tradeoff , 1970, IEEE Trans. Inf. Theory.

[6] R. Rosenfeld,et al. Two decades of statistical language modeling: where do we go from here? , 2000, Proceedings of the IEEE.

[7] Stanley F. Chen,et al. Building Probabilistic Models for Natural Language , 1996, ArXiv.

[8] F ChenStanley,et al. An Empirical Study of Smoothing Techniques for Language Modeling , 1996, ACL.

[9] Eric Brill,et al. Beyond N-Grams: Can Linguistic Sophistication Improve Language Modeling? , 1998, COLING-ACL.

[10] Fumitaka Kimura,et al. Impact of lexicon completeness on city name recognition , 2002, Proceedings Eighth International Workshop on Frontiers in Handwriting Recognition.