基于深度学习的图像文本切分与识别 Image text segmentation And recognition based on deep learning 内容摘要 自从 1929 年德国科学家提出 OCR 的概念，各个国家就开始对此展开研究，OCR 全称 Optical Character Recognition，即光学字符识别一开始专家们并没有对字母、单词、文字、字形等进行研究，就像电话还没有被发明之前一样，人们觉得这是天方夜谭，研究要从基础开始，因此，但是的人们是从最简单的 10 个数字（0-9）开始的。由于历史原因，中文识别起步较晚，并且由于汉字字形与由字母组成的英文、法文等不同，汉字字形各异，组织结构复杂，机器寻求其中的规律比较困难，常常会因为偏旁部首出现切分错误，要精准地识别并不容易，可以说是相当有挑战性的。随着信息化水平不断的提升，图像时代已经越来越近，这是一件必然的事情，当我们拥有足够的科技，足够的能力，印刷文化将会被新的视觉文化所取代，识别技术的发展势不可挡，我们能很明显地感受到身边相关的技术，百度、谷歌等都有相关应用。在字符识别方面，可选择的有谷歌 Tesseract、百度 API、传统的字符特征提取、模板匹配法以及基于深度学习下的 CNN 字符识别。本文使用模板匹配法以及基于深度学习下的 CNN 字符识别相互结合的方法。关键词：OCR 中文文本识别 卷积神经网络 文本检测Abstract Since 1929, German scientists put forward the concept of OCR, Countries began to study it, and OCR full name optical character recognition, At the beginning, experts didn't study letters, words, characters, glyphs, etc., just like before the telephone was invented, people thought it was a fantasy, and the research should start from the foundation, so, people started from the simplest 10 numbers (0-9).For historical reasons，Chinese recognition started late, and due to the differences between Chinese characters and alphabetic English, French, etc, Chinese characters have different shapes and complicated organizational structure, so it is difficult for machines to find the rules among them, There are often segmentation errors due to the radicals, it is not easy to...
