本文介绍了中文文本检测和识别的任务，以及ECCV 2016上提出的基于Faster R-CNN的Connectionist Text Proposal Network(CTPN)方法。CTPN利用双向LSTM网络，针对文本检测的特性，提高了检测准确性。此外，文章讨论了Faster R-CNN在文本检测中的应用，以及Tesseract开源库在文本识别中的局限性。
原文站点：https://senitco.github.io/2017/03/03/text-detection-recognition/
The task of Chinese text detection is to localize the regions in a 2D image which contain Chinese characters. The task of Chinese text recognition is, given the localized regions including text, to convert each region into machine-encoded text. It is an important technique for understanding text information in 2D images, and many other applications such as text-to-speech, machine translation, text mining, etc.
Course
Image Analysis and Understanding
Project
Chinese Text Detection and Recognition
Description
The task of Chinese text detection is to localize the regions in a 2D image which contain Chinese characters. The task of Chinese text recognition is, given the localized regions including text, to convert each region into machine-encoded text. It is an important technique for understanding text information in 2D images, and many other applications such as text-to-speech, machine translation, text mining, etc.
Deadline
April 11, Tuesday, 2017
Overview
图像文本一般分为两大类，图形文本(Graphic text)和场景文本(Scene text)，图形文本通常是指覆盖在图像上的机器打印的文本，如图片水印、视频字幕等；而场景文本则是自然场景中物体上的文本，如墙壁上的标语、衣服上的文字等。最近大部分的研究都集中在自然场景中文本检测与识别上，由于受到场景复杂、光照变化、成像随意等因素的影响，这一研究任务具有较大
最低0.47元/天 解锁文章
520
被折叠的 条评论
为什么被折叠?
到【灌水乐园】发言
