基于手绘草图的图像检索研究
发布时间:2018-03-22 22:02
本文选题:基于内容的图像检索 切入点:草图检索 出处:《大连理工大学》2013年硕士论文 论文类型:学位论文
【摘要】:随着互联网技术的发展,数字图像的数量急剧增长,基于内容的图像检索技术引起了国内外学者广泛的关注并取得了显著的研究成果。近几年随着当今社会触摸屏设备如平板电脑和智能手机等的普及,人们开始将关注的重点转移到基于手绘草图的检索技术中。在触摸屏设备的帮助下,各个年龄段、各种绘画水平的人都可以轻松绘制出浮现在脑海中的物体,进而通过手绘的线条图在大量图片库中找到与之形状类似的图像。 基于草图的图像检索这一概念最早在20世纪80年代被提出,但是之后却一直进展较慢,主要是因为手绘草图中线条的多变性和不确定性使得线条的特征表示、特征匹配以及适合大规模数据库的索引结构的建立等方面充满了困难和挑战。2010年微软亚洲研究院提出的可以不依赖关键字,只根据草图中物体线条的特征匹配在大规模图片数据库上进行实时检索的草图搜索系统MindFinder再一次引发了人们对草图搜索的研究热情。 本文提出的基于手绘草图的图像检索系统采用“词袋模型”来表示草图,每幅草图均可表示为与视觉字典中的单词相关的直方图。在特征提取过程我们使用的局部特征描述子是经过改进的梯度方向直方图特征--基于草图梯度场的梯度方向直方图(GF-HoG),该特征能够有效地表示由线条构成的草图;在构建视觉字典时我们采用分层K-means聚类算法,该聚类算法与传统的K-means聚类算法相比能获得更精确的聚类结果。最后通过比较输入草图与图像库中图像之间的余弦相似性实现库内检索过程,通过多类SVM分类器可以得到输入草图所属类别的关键字,将关键字送至搜索引擎能够实现在线检索。我们在Eitz提供的手绘草图数据库和Microsoft Office文档中提供的形状图数据库中进行了一系列实验,实验结果表明本论文提出的手绘草图检索算法与MindFinder和Eitz的草图检索算法相比效果更加显著。
[Abstract]:With the development of Internet technology, the number of digital images has increased dramatically. Content-based image retrieval technology has attracted wide attention from scholars at home and abroad and achieved remarkable research results. In recent years, with the popularity of touch screen devices such as tablets and smart phones, People are starting to shift their focus to hand-sketch-based retrieval technology. With the help of touchscreen devices, people of all ages and levels of painting can easily draw objects that come to mind. And then through the hand-drawn line map in a large number of photo library to find the shape of the image. The concept of Sketch-based Image Retrieval was first proposed in the 1980s, but the progress has been slow since then, mainly because of the variability and uncertainty of the lines in the hand-drawn sketches. Feature matching and the creation of index structures suitable for large-scale databases are fraught with difficulties and challenges. The sketch search system (MindFinder), which only performs real-time retrieval based on the feature matching of objects and lines in sketches, has once again aroused people's enthusiasm for sketch search. This paper presents an image retrieval system based on hand-drawn sketches, which uses a "word bag model" to represent sketches. Each sketch can be represented as a histogram associated with a word in a visual dictionary. The local feature descriptor we use in the feature extraction process is an improved gradient direction histogram feature-a gradient based on the gradient field of the sketch. Direction histogram (GF-HoG), which can effectively represent sketches composed of lines; In constructing visual dictionary, we adopt hierarchical K-means clustering algorithm. Compared with the traditional K-means clustering algorithm, this clustering algorithm can obtain more accurate clustering results. Finally, the retrieval process in the database is realized by comparing the cosine similarity between the input sketches and the images in the image database. The keywords of the category to which the input sketch belongs can be obtained by the multi-class SVM classifier. Sending keywords to search engines enables online retrieval. We have done a series of experiments in the sketching database provided by Eitz and the shape map database provided in Microsoft Office documents. Experimental results show that the proposed hand-drawn sketch retrieval algorithm is more effective than that of MindFinder and Eitz.
【学位授予单位】:大连理工大学
【学位级别】:硕士
【学位授予年份】:2013
【分类号】:TP391.41
【参考文献】
相关期刊论文 前9条
1 王千;王成;冯振元;叶金凤;;K-means聚类算法研究综述[J];电子设计工程;2012年07期
2 金微;陈慧萍;;基于分层聚类的k-means算法[J];河海大学常州分校学报;2007年01期
3 陈剑峗,老松扬,吴玲达;基于内容的图像检索的发展最新趋势[J];计算机工程与应用;2002年10期
4 陈峗;沈一帆;;基于词汇树的图片搜索[J];计算机工程;2010年06期
5 唐发明,王仲东,陈绵云;支持向量机多类分类算法研究[J];控制与决策;2005年07期
6 苟博;黄贤武;;支持向量机多类分类方法[J];数据采集与处理;2006年03期
7 向培素;;聚类算法综述[J];西南民族大学学报(自然科学版);2011年S1期
8 苏晓珂,虎晓红,兰洋;基于内容的图像检索技术综述[J];信阳师范学院学报(自然科学版);2005年04期
9 马继红;师军;;基于内容的图像检索技术研究[J];郑州轻工业学院学报(自然科学版);2009年04期
相关会议论文 前1条
1 龚健;费广正;石民勇;曹玮;;基于手绘草图轮廓检索的简笔绘图系统[A];'2008系统仿真技术及其应用学术会议论文集[C];2008年
相关硕士学位论文 前1条
1 夏荣进;基于形状上下文的图像内容检索方法研究[D];华中科技大学;2011年
,本文编号:1650575
本文链接:https://www.wllwen.com/kejilunwen/sousuoyinqinglunwen/1650575.html