Application of Fuzzy Matching Algorithms for Doctors Handwriting Recognition

Abstract

Doctor's handwritten prescriptions are often known to be indecipherable. Uncertainty in medical terms can have dire consequences. A method to effectively recognize medicine names written in doctor's handwriting is proposed in this paper. A corpus of 600 images is compiled with the help of multiple doctors. An exhaustive list of 50 medicines is used for the same. Recognition is performed using the Convolutional Recurrent Neural Network (CRNN) - Connectionist Temporal Classification (CTC) model which results in 93.3 % accuracy. In order to deal with errors produced in the recognized text, edit distance methods are further implemented and analyzed. Damerau-Levenshtein distance method is deemed to be the most suitable, yielding a well-grounded system for medicine name recognition.