A Lexicon and Depth-Wise Separable Convolution Based Handwritten Text Recognition System

Abstract

AbstractCursive handwritten text recognition is a challenging research problem in pattern recognition. The current state-of-the-art approaches include models based on convolutional recurrent neural networks and multi-dimensional long short-term memory recurrent neural network techniques. These methods are highly computationally extensive as well model is complex at the design level. In recent studies, a combination of convolutional neural networks and gated convolutional neural networks based models demonstrated less number of parameters in comparison to convolutional recurrent neural networks based models. In the direction to reduced the total number of parameters to be trained, in this work, we have used depthwise separable convolution in place of standard convolutions with a combination of gated-convolutional neural network and bidirectional gated recurrent unit to reduce the total number of parameters to be trained. Additionally, we have also included a lexicon-based word beam search decoder at the testing step. It also helps in improving the overall accuracy of the model. We have obtained 3.84% character error rate and 9.40% word error rate on IAM dataset, 3.15% character error rate and 11.8% word error rate on RIMES dataset and 4.88% character error rate and 14.56% word error rate in George Washington dataset respectively.KeywordsDepthwise separable convolutionCursive handwritten text line recognitionWord beam searchDeep learning