基于边缘信息的文本区域提取算法的研究-计算数学专业论文.docxVIP

AB AB STRACT ABSTRACT：Text is the important content information in images and video f／ames．It’S very important for image comprehending，retrieving and browsing that automatic locating and extracting the text regions in image and video frame，then recognizing characters by OCR technique，and last extracting the text from images． In this thesis，we study on text region extraction in color images．It includes five parts：multi。scale analysis，color edge extraction，morphologic processing，connected -component analysis and text regions extraction．The method proposed in this thesis improves on the present scene text and graphic text region extraction method，and achieves rather extraction accuracy． The size of text in image varies a lot，to extract different size text，make the image into three resolutions：1／1，1／2，1／4 by multi—scale analysis．The same text region extraction algorithm is applied for different resolution sub-image．Due to the direction of text distribution is various，to extract text region more accurately,a morphologic processing algorithm based on the direction of text distribution is proposed in this thesis． Its basic idea is that to classify the text distribution into horizontal，vertical，upward， downward direction．As for different direction of text distribution,we use different morphologic processing algorithm．In the processing of connected—component analysis， we discuss a variety of situations when we get the width and regularization of connected -component．Moreover,we analyze in detail．So the result of detection is ideal． At present，there are many algorithms proposed to as for some specific sample set． Once the sample is changed，the results谢th these algorithms are not ideal．The method proposed in this thesis is not as for some specific set，SO the range of its applicability is wide relatively． KEYWORDS：Text region extraction；Multi·scale analysis；Color edge extraction； Morphology；connected—component analysis CLASSNO：TP75 1．1 学位论文版权使用授权书本学位论文作者完全了解北京交通大学有关保留、使用学位