Publication

Abstract : The goal of this project is to accurately digitize a variety of handwritten texts in order to address the problem of handwriting recognition (HR). Because handwritten scripts are inherently complex and handwriting styles vary widely existing HR systems frequently face challenges. Convolutional neural networks (CNN) are used for accurate feature extraction and recurrent neural networks (RNN) with Long Short-Term Memory (LSTM) cells are used for effective sequence learning in our sophisticated HR system. Connectionist Temporal Classification (CTC) improves the system even more by facilitating efficient alignment of text sequences. With the addition of AutoCorrect features our study shows a noteworthy improvement in HR accuracy with a character error rate (CER) of 8% and a word error rate (WER) of 12%. This is a significant improvement over the initial 15 percent CER and 20 percent WER before AutoCorrect was integrated. The robustness and efficiency of the system were demonstrated by these metrics which underwent extensive testing on a complex and diverse dataset.