Research On Scene Character Recognition Technology In Image

With the development of the Internet and the popularity of image input devices,images have drawn more and more attention in people's lives.In addition to more vividly displaying things,Images contain more information than simple words,the advantage of obtaining textual information from images is that textual information associated with images can be obtained.By further processing of textual information,such as Internet retrieval,more text messages can be gotten.But getting textual information from images has still been a troubling topic.In order to be able to extract text information from a variety of complex images,a variety of text recognition theories have been put forward and achieved some results in practice,in recent years,scene text recognition technology has been favored by computer vision community researchers,and has been in research and development among.Based on the existed scene text recognition algorithms,this thesis studies the theory and practice of scene text recognition technology in-depth.In the text recognition method,this thesis analyzes the traditional text recognition technology and non-traditional text recognition theory methods,and introduces two popular text recognition algorithms,that is based on the MSER algorithm and based on the SWT algorithm,an improved algorithm based on SWT is introduced;In the method of text extraction,this thesis analyzes the traditional and non-traditional methods of text extraction,and introduces two kinds of popular text extraction algorithms based on the Otsu algorithm and the watershed algorithm.An improved Based on Otsu algorithm is introduced;In the method of Deep Learning on text recognition,this thesis analyzes the general framework and process of deep learning,and introduces the popular text scene recognition algorithm based on CTPN,finally,a text recognition system based on CTPN + CRNN is introduced.With the practice simulation,this thesis based on SWT improves the process algorithm to speed up the recognition speed by 38%.The practice results show that the improved algorithm accords with the actual requirements and has achieved good results.In the method of text extraction,Otsu algorithm make it possible to have a better recognition effect.Practice shows that the improved Otsu algorithm is better than the traditional Otsu algorithm Although sacrificed for some speed.In the method of deep learning about text recognition,based on the CTPN algorithm,this thesis designs a scene text recognition system of CTPN + CRNN,the system of the accuracy and recall rate performed well,the practice of scene text recognition practice has some reference.