项目名称: 基于局部特征的自然场景下文字定位和识别研究
项目编号: No.61201384
项目类型: 青年科学基金项目
立项/批准年度: 2013
项目学科: 电子学与信息系统
项目作者: 周异
作者单位: 上海交通大学
项目金额: 24万元
中文摘要: 自然场景中的文字定位和识别在互联网信息理解/智能交通等众多领域具有重要的应用价值,但该研究面临复杂背景、低图像质量以及文字变形等诸多挑战。本课题研究包括:(1)结合图像检索/物体识别领域的研究方法和成果,提出LCR模型,并基于该模型设计基于局部特征的文字定位和识别总体架构,设计几何约束算法和样本库构建算法;(2)研究文字结构及统计特性,提出一种面向文字的局部不变性特征检测及特征描述算法,并研究利用局部特征改进文字定位/字体识别的方法。目前基于局部特征的方法尚处于起步阶段,该方法基于局部特征本身具有不变性特征(旋转不变性、尺度不变性、仿射不变性、灰度不变性等),在解决复杂背景与布局、低质量及文字变形的挑战方面展现了巨大的研究潜力。本课题将结合申请人已有的研究成果,形成文字定位和识别系统的研究平台,为自然场景中文字定位和识别提供理论依据和实践基础。
中文关键词: 中文文字识别;自然场景;卷积神经网络;局部特征;
英文摘要: Our goal is to read text from an image in natural scenes. There are many applications for such a technology, for example, recognizing sign from natural scenes, license plate recognition, image and video search engine and web mining. However, low image quality, complex background, deformation and variations of text make these problems challenging. Our research includes: (1) Propose a LCR model based on the research works of object recognition and image retrieval, design a framework of text localization and recognition using local features, and design algorithms of geometric verification and template images generation. (2) Propose algorithms of local features detector and descriptor for text, and develop methods of text localization and font recognition. Although there are still many works are needed toward a mature application for the local features method. Local features which are distinctive and robust to noise, complicated background, and many kinds of geometric and photometric deformations. Moreover,as our prior works show, local features matching could be potentially extended to text recognition problems. Our research work will build a platform of text localization and recognition using local features, and will provide basis of theoretical and practical for research of text localization and recognition in
英文关键词: Chinese Character Recogniton;Natural Scene;CNN;Local Features;