Application Research Of Recurrent Neural Network In Optical Character Recognition

Optical character recognition is a technique to detect and recognize text in images.Compared to traditional computer vision algorithms,convolutional neural network and recurrent neural network are more robust to different backgrounds and shapes and they are the most attractive field recently.In this paper,we study the applications of recurrent neural network in optical character recognition and focus on text detection and recognition in natural scenes.We propose a textline expanding and chaining algorithm based on a recurrent neural network,a word segmentation algorithm based on a recurrent neural network and a text recognition algorithm based on a attention model.Finally we combine our models and a text detection algorithm based on a convolutional neural network to build a optical character recognition system.Experiments show that our textline expander can effectively retrieve undetected characters,our word segmentation model can do segmentation under complex backgrounds and our recognition model and OCR system is close to state of the art on some related datasets.