图片视频中叠加文字提取识别技术研究

项目名称： 图片视频中叠加文字提取识别技术研究

项目编号： No.60873087

项目类型： 面上项目

立项/批准年度： 2009

项目学科： 轻工业、手工业

项目作者： 王伟强

作者单位： 中国科学院大学

项目金额： 30万元

中文摘要： 准确提取识别嵌入在图片视频中的文字对于帮助计算机理解多媒体信息的语义具有重要的研究意义与应用前景。围绕该目标本项目将系统地研究涉及的各种关键技术，包括复杂背景中的文字检测、文字分割，以及非理想分割状况下的汉字识别，并从系统的角度优化各项技术的算法。在文字检测与分割方面，我们将重点研究笔画特征的描述表示，并希望将文字检测与分割算法有机结合在一起，通过特征信息共享优化整体算法性能。非理想分割结果下的汉字识别方法研究目前在国际上还是一个空白，我们以此为应用背景研究抗干扰的汉字识别特征，以及研究面对不确定性扰动的样本如何利用支持向量机SVM来建模求解大数目类别(6000多类)的分类问题。本项目的研究内容不仅与实践应用紧密相关，同时项目潜在的研究成果对丰富学科基础理论(支持向量机SVM求解大数据类分类问题)也具有重要价值。

中文关键词： 叠加文字； 文字检测； 文字提取； 文字识别

英文摘要： It is very significant to accurately extract embedded text in images and videos for computers to automatically understand the sementics of multimedia information. To this end, the project will systematically investigate various key technologise, including text detection, text segmentation in complex background, and robust character recognition under nonideal segmentation; at the same time the technologies will be optimized from the system perspective. For text detection and segmentation, we focus on the effective description and representation of strokes, and expect to integrate the detection and segmentation precedures together to share some feature information so as to improve the overall system performance. To our knowledge, few researches are reported on character recognition under nonideal segmention results. For the application, we investigate the related robust recognition features, and how to use SVM to model the problem of a large number of categories with noisy samples. The research contents of the project are tightly related with practical application, and at the same time the potential research results are very valuable for enriching the fundamental theory of machine learning ( the classification issue of using SVM to model a large number of categories with noisy samples).

英文关键词： embedded text；text detection; text extraction; character recognition