Publication

Abstract : In the present scenario, Handwriting Recognition (HWR) plays a very important role as due to current pandemic situations most of the exams are conducted in online mode. In HWR the device translates the user’s handwritten characters or words into readable by a computer system. The problem with HWR is there are different types of handwriting and different styles of writing each character which makes it difficult to analyze each letter uniquely and correctly also the major problem with HWR is when people write there is no specific font and font sizes are taken care which also plays important role in recognizing the letters. The paper presents an accuracy comparison system of spelling correction in Handwritten OCR data using four neural models BERT, SC-LSTM, CHAR-CNN-LSTM, CHAR-LSTM-LSTM. In task, sequence matcher algorithm is used for computation of similarity score between input text data and outputs of the different neural models at each iteration. It is observed that the BERT model gives the highest accuracy of 71.4% followed by SC-LSTM with 69.12%, CHAR-CNN-LSTM with 67.80% and CHAR-LSTM-LSTM with 69.34%.