Handwritten word recognition with deep learning

Deep learning, the study of multi-layer artificial neural networks, had a significant impact on a number of artificial intelligence domains, including handwriting recognition. In this work, several insights from the domains of deep learning and handwriting recognition were focused. This paper seeks to classify an individual handwritten word so that handwritten text can be translated to a digital form. To complete this objective, two basic strategies were employed: firstly, character segmentation and secondly, classifying the words. For the former, to train a model that can correctly categories words, Recurrent Neural Network (RNN) was used with various architectures. For the latter, bounding boxes were created for each character using Long Short Term Memory Networks (LSTM) with convolution. After segmenting the characters, data were fed to a Convolution Neural Network (CNN) for classification. The IAM handwriting dataset was primary source of training the data for handwriting recognition system. Over 96,000 samples of handwritten text are included in this dataset. Based on the classification and segmentation outcomes, each word was reconstructed.