三維空間實景文字之偵測、切割與辨識; 3D Text Detection, Segmentation, and Recognition in 3D Scenes

NCU Institutional Repository > 資訊電機學院 > 資訊工程學系 > 研究計畫 > Item 987654321/47133

jsp.display-item.identifier=請使用永久網址來引用或連結此文件: http://ir.lib.ncu.edu.tw/handle/987654321/47133

题名:	三維空間實景文字之偵測、切割與辨識;3D Text Detection, Segmentation, and Recognition in 3D Scenes
作者:	范國清
贡献者:	資訊工程系
关键词:	資訊科學--軟體
日期:	2010-08-01
上传时间:	2011-07-14 09:59:22 (UTC+8)
出版者:	行政院國家科學委員會
摘要:	文字辨識(OCR)技術是影像處理及圖形識別最成功的應用案例之一，然而傳統之文字辨識卻無法解決日益急迫的三維文字辨識問題。因為全球化所造成各國間交流越趨頻繁，各種語文間翻譯的需求量越來越高，但翻譯人員並非隨處皆有，只能借助自動化翻譯機器的協助，但三維文字的輸入與辨識本身就是一個極大的問題與挑戰，除非是翻譯文字檔案，否則根本不易於進行出國時日常生活隨處可見的三維場景中招牌、菜單、路標等外國文字上的輸入，此時問題的核心，就在於如何進行三維場景中文字的偵測、切割與識別。在本研究計畫中；第一年我們將建立一套web-camera 平面文字辨識系統，這套系統將提供使用者中英文辨識的功能。第二年將建構一套規則立體表面上文字辨識的模型。第三年將嘗試在第二年的基礎上更往前一步，建構一套不規則曲面上文字辨識的模型，以將辨識範圍由規則曲面擴展至非規則曲面上。本計畫擬建立一套以電腦視覺為基礎之智慧型的文字辨識系統，希望能夠使用現今科技帶來的便利性與機動性，來提供使用者一種隨身翻譯秘書的功能 Optical Character Recognition (OCR) is admitted as one of the most successful and mature applications in the areas of image processing and pattern recognition. However, the recognition of 3D characters in 3D scenes is still a challenging problem which is much difficult than traditional 2D character recognition problem. Due to the internationalized and frequent visit between different countries, the demand for the real-time translation of different languages gradually becomes a must. However, the input of 3D texts, such as road signs, campaigns, and manus, are infeasible using traditional 2D scanning devices. The problems encountered are then the solving of those unexpected effects incurred by the 3D image capturing devices in order to successfully extract the characters embedded in the real scene environments. In this proposal, we plan to build a web-camera based character recognition system which can recognizes 3D English/Chinese characters captured by 3D capturing devices. Name card like planar objects are the simplest 3D texts to be resolved in the first year of this project. In the second year, we plan to extend the objects to regular non-planar objects and recognize the embedded characters accordingly. As to the third year, the objects to be dealt with are further lifted to irregular non-planar objects. The handling of non-uniform lighting, perspective distortion, and camera vibration are the challenging issues to be conquered in the execution of this project which have seldom been addressed or even ignored in traditional OCR/DA research. With the ongoing of this project, we hope to develop a personal portable assistant to resolve the machine translation problem. 研究期間：9908 ~ 10007
關聯:	財團法人國家實驗研究院科技政策研究與資訊中心
显示于类别:	[資訊工程學系] 研究計畫

文件中的档案:

档案	描述	大小	格式	浏览次数
index.html		0Kb	HTML	486	检视/开启

在NCUIR中所有的数据项都受到原著作权保护.

社群 sharing

数据加载中.....