Physical Sciences and Mathematics Commons^™

Post Processing Of Optically Recognized Text Using First Order Hidden Markov Model, Spandana Malreddy

UNLV Theses, Dissertations, Professional Papers, and Capstones

In this thesis, we report on our design and implementation of a post processing system for Optically Recognized text. The system is based on first order Hidden Markov Model (HMM). The Maximum Likelihood algorithm is used to train the system with over 150 thousand characters. The system is also tested on a file containing 5688 characters. The percentage of errors detected and corrected is 11.76% with a recall of 10.16% and precision of 100%

Post Processing Of Optically Recognized Text Via Second Order Hidden Markov Model, Srijana Poudel

UNLV Theses, Dissertations, Professional Papers, and Capstones

In this thesis, we describe a postprocessing system on Optical Character Recognition(OCR) generated text. Second Order Hidden Markov Model (HMM) approach is used to detect and correct the OCR related errors. The reason for choosing the 2nd order HMM is to keep track of the bigrams so that the model can represent the system more accurately. Based on experiments with training data of 159,733 characters and testing of 5,688 characters, the model was able to correct 43.38 % of the errors with a precision of 75.34 %. However, the precision value indicates that the model

introduced some new errors, decreasing …