An Intelligent Automatic Tranlation System of Printed Document into Braille at high accuracy

印刷文件高精度自动盲文翻译系统

• 批准号: 04508001
• 负责人: KIMURA Masayuki
• 金额: $ 28.03万
• 依托单位: Japan Advanced Institute of Science and Technology, Hokuriku
• 依托单位国家: 日本
• 项目类别: Grant-in-Aid for Developmental Scientific Research (A)
• 财政年份: 1992
• 资助国家: 日本
• 起止时间: 1992 至 1994
• 项目状态: 已结题
• 来源: https://kaken.nii.ac.jp/grant/KAKENHI-PROJECT-04508001/
• 关键词: Automatic transcription into braille; Printed character recognition; Document understanding; Visually-disabled person; Document reader; User-interface; Reading aids; 墨字文書; 点字文書; 文書画像のレイアウト; 文字認識; 点訳システム; 文字認識を導入した自動点訳; 日本語英語自動点訳; 文書画像の領域理解

The developed system mainly consists of two modules. The first one is the OCR (Optical Character Recognition unit) module which performs the analysis and recognition of scanned text, and the second module outputs the result in the Braille text form.The main processing unit of the OCR module includes a UNIX workstation and a character recognition device which has been specially developed for this project.In order to fulfilll the various demands of a visually impaired person, several kinds of texts like magazine (weekly magazine), conference journal, text books, technical books, paperbacks, and other practical documents has been taken into account. For analyzing the structure of a text, page numbers, position of diagrams/tables, captions and ruby characters has been considered for recognition. During experiments the recognition rate for line and diagram detection was 97.4respectively. However, because of the variation of text formats, the recognition rate for detecting caption, page number and foot notes was widely varied.The developed character recognition unit is able to process 25 to 28 characters per second and it has been confirmed that the present processing speed is 5 to 8 times faster if compared to the method where the processing is done by employing software. By utilizing dictionary and by modifying the algorithm the correct output for Braille text is improved to 99.5The interface design was based on on a cognitive model and its effectiveness is verified by a number of impaired users.Finally it is necessary to perform research for improving the response in respect to key operations and evaluate the system with the help of a number of visually impaired user.

开发的系统主要由两个模块组成。第一个是OCR（光学字符识别单元）模块，对扫描文本进行分析和识别，第二模块将结果以盲文文本形式输出。OCR模块的主处理单元包括UNIX工作站和专门为此项目开发的字符识别设备。为了满足视障人士的各种需求，已经考虑了几种文本，如杂志（周刊杂志）、会议杂志、教科书、技术书籍、平装本和其它实用文档。为了分析文本的结构，考虑了页码、图表/表格的位置、标题和红宝石字符的识别。在实验中，直线和图形的识别率分别为97.4%。然而，由于文本格式的变化，检测标题、页码和脚注的识别率变化很大，所开发的字符识别单元每秒能够处理25至28个字符，已经证实，如果与采用软件进行处理的方法相比，目前的处理速度快5至8倍。通过使用字典和修改算法，盲文文本的正确输出率提高到99.5%，界面设计基于认知模型，并通过多个视障用户验证了其有效性，最后需要进行研究以改善对关键操作的响应，并通过多个视障用户对系统进行评估。

项目成果

木塚泰弘: "視覚障害情報処理技術 1993年の特徴と1994年の課題" 視覚障害. 34-38 (1994)

Yasuhiro Kizuka：“视觉障碍信息处理技术：1993 年的特征和 1994 年的挑战”视觉障碍 34-38 (1994)。

木村正行,勝山裕: "An Intelligent Automatic Translation System From a Printed Document in Japanese into Braille" IISF International Symposium on Computers as Our Better Partners. (論文集). 150-151 (1944)

Masayuki Kimura、Yutaka Katsuyama：“从日语印刷文档到盲文的智能自动翻译系统”IISF 国际计算机研讨会作为我们更好的合作伙伴（论文集）150-151（1944 年）。

米田政明: "多カテゴリ認識問題の正解率と期待値に関する一考察" 第16回情報理論とその応用に関するシンポジュウム. 633-636 (1993)

米田正明：“多类别识别问题的准确率和期望值的研究”第16届信息论及其应用研讨会633-636（1993）。

H.Aso: "Thinning Algorithm Suitable for Parallel Processing" Trans.IEICE Japan (D-II). J76-D-II,9. 2148-2150 (1993)

H.Aso：“适合并行处理的细化算法” Trans.IEICE Japan (D-II)。

勝山裕,木村正行: "知的自動点訳システム" 視覚障害者ワークショップACM,IISF国際シンポジウム. 4 (1994)

Yutaka Katsuyama、Masayuki Kimura：“智能自动盲文翻译系统”视障 ACM 研讨会，IISF 国际研讨会 4（1994 年）。