Handwritten Chinese character recognition method based on full-convolution recursive network

Abstract

The invention discloses a handwritten Chinese character recognition method based on a full-convolution recursive network. The method comprises the following steps: a path integral layer converts online handwriting information into a corresponding offline feature image; a full-convolution network extracts the high-dimensional abstract expression of the offline feature image and generates a corresponding response diagram; a multilayer bidirectional recursive network identifies each frame of the response diagram and outputs a probability distribution about a character set; a transcription layer uses a dynamic programming algorithm of forward calculation and reverse gradient spread to make a whole handwritten Chinese character recognition model be trained directly based on text data; and a language model is post-processed. The method has different description ability for the original online handwriting information. Under the condition that a handwritten Chinese text is not pre-segmented, an input sequence of arbitrary length can be received, and a corresponding output sequence can be output. The method has strong overall performance. By embedding a language model through a beam search method to decode a full-convolution recursive network, the recognition rate is improved.

Description

Technical field [0001] The present invention relates to a technology for recognizing handwritten Chinese character text on an online handwritten document input by a computer user's handwriting to a computer, in particular to a handwritten Chinese character text recognition method based on a full convolution recursive network. Background technique [0002] Handwritten Chinese character recognition is a challenging problem in today's world, and it has received close attention from many researchers. Large character sets, diverse handwriting styles and character connection problems are the main problems encountered in handwriting Chinese characters. In recent years, handwritten Chinese character recognition has made certain breakthroughs, but traditional methods based on over-segmentation still fail to overcome the problem of correcting character segmentation errors, and the recognition accuracy is not high. Summary of the invention [0003] In order to overcome the shortcomings of t...

AI Technical Summary

Problems solved by technology

In recent years, some breakthroughs have been made in the recognition of handwritten Chinese characters, but the traditional method based on over-segmentation still fails to overcome the problem of correcting the wrong segmentation of characters, and the accuracy of recognition is not high

Images