Text Identification from Handwritten Data using Bi-LSTM and CNN with FastAI

Abstract

Text extraction is critical for any analysis in a document processing system. Text extraction is the process of recognizing text data from an image. The handcrafted elements used by traditional handwriting recognition systems require a lot of prior knowledge. Convolutional approaches can be used to train optical character recognition (OCR) systems, although doing so requires a lot of training data. Deep learning approaches are the main focus of handwriting recognition research, which has recently produced ground-breaking results. But the exponential expansion of handwritten text and the accessibility of vast computational power need an improvement in predictive performance and more study. To enable the automatic extraction of distinguishing features from handwritten characters and phrases, Convolutional Neural Networks (CNNs), a subset of Deep Learning technology, are especially adept at comprehending the structure of handwritten letters and phrases. The disadvantages of this approach include increased time and resource requirements. The proposed design is based on CNN with Bi-LSTM, is used to identify the text from the handwritten images.