Handwritten Text Recognition using Deep Learning Methods

Offline handwritten text recognition has been widely utilized in various fields including historical document analysis. Deep learning techniques have demonstrated their effectiveness in digitizing handwritten text as each technique is precisely designed to tackle a specific task or solve a particular problem. In this article, we use convolutional neural network for extracting distinct character features and a recurrent neural network for handling character combinations within sequential data. By combining these models, we create a hybrid deep neural network consisting of three CNN layers followed by a bidirectional LSTM layer. This architecture effectively encodes input images and generates character probability matrices with which the connectionist temporal classification operation computes the loss function. Extensive experimentation with various parameter values allowed us to optimize our model, which we evaluated on the IAM dataset, yielding a reasonably low error rate.