Design and Implementation of a Hybrid Deep Learning Framework for Handwritten Text Recognition

Text recognition technology has seen significant advancements in recent years, particularly with the use of Optical Character Recognition (OCR) to evaluate computer-generated text. However, there is much more work to be done in the field of Handwritten Text Recognition (HTR). The challenges posed by handwritten text, such as significant variations in strokes across writers, the vast variety of handwriting styles, human error and damages to the paper, present substantial difficulties in accurately identifying and recognizing handwritten alpha-numeric data. To address these challenges, we proposed a deep learning method that combines long short-term memory (MD-LSTM) and convolutional neural networks (CNN) architectures. This model can identify numbers and characters of the English language from input images. Based on MNIST dataset, bidirectional recurrent neural networks were used to construct the output sequence which was developed using the TensorFlow Framework. The accuracy for alphabets, numbers and alpha-numeric texts are 95.2%, 94.9% and 94.7% respectively. The mean character match index is computed to be 93.2%. The proposed model can substantially boost HRTs precision and efficiency making it more accessible.