Character recognition method based on deep learning

Abstract

The invention discloses a character recognition method based on deep learning. The method includes the construction phase of a spatial transform layer and the construction and training phase of a deep convolutional neural network. The spatial transform layer consists of three parts: a positioning network receives a feature map as an input, enables the feature map to pass through a series of hidden layers, and then outputs a parameter of spatial transform, wherein the parameter is to be used for the feature map; a grid generator generates sampling grids by using the parameter generated in the first part; and a sampler takes the feature map and the sampling grids as inputs, samples the feature map at grid points, and finally obtains an output feature map result. The spatial transform layer can be differentiated. By the spatial transform layer, image data can be spatially processed in the network so that the network can learn the invariance of the space distortion and avoid the need for manually generating a large number of deformed samples in traditional convolutional network training. In addition, by building deeper volumes and neural networks, a better recognition effect is realized on a large variety of Chinese characters.

Description

technical field [0001] The invention belongs to the field of character recognition in pattern recognition, and more specifically, relates to a method of character recognition based on deep learning. Background technique [0002] With the continuous development of modern science and technology and the widespread popularization of the Internet, we have to come into contact with massive information resources presented in various forms every day, especially in our daily life, study and work, it is often unavoidable to deal with a large number of Text information and enter it into the computer. Therefore, how to quickly and accurately input these text information into various electronic devices such as computers has become an urgent problem to be solved. Optical Character Recognition (OCR for short) refers to a technology that automatically extracts the text in the picture with the help of machine equipment and converts it into text that can be edited by the machine. [0003] G...

AI Technical Summary

Problems solved by technology

[0007] However, there are deficiencies in traditional Chinese font recognition methods. Because of the complexity of Chinese characters, feature extraction methods cannot deal with changing Chinese character shapes. Feature point extraction methods require manual experts to define important feature point positions. Moreover, for those feature points The importance cannot be given a unified standard, resulting in a lower accuracy of text recognition

Images