CASIA OpenIR  > 毕业生  > 硕士学位论文
交互式图像目标分割及检索系统
Alternative TitleInteractive Foreground Extraction and Image Retrieval System
仇浩文
Subtype工学硕士
Thesis Advisor肖柏华
2015-05-27
Degree Grantor中国科学院大学
Place of Conferral中国科学院自动化研究所
Degree Discipline模式识别与智能系统
Keyword交互式分割 目标提取 图像检索 Interactive Segmentation Foreground Extraction Image Retrieval
Abstract随着带拍照功能的移动智能终端的广泛普及和互联网技术的迅猛发展,我们已经步入“移动互联网时代”和“读图时代”,拍照检索日渐成为人们对真实世界检索的重要入口之一。 在拍照检索中,人们往往更多关注的是图片中某个感兴趣的目标,如商品、场所、标志等,而不是整个图片的内容。这给基于内容的图片检索研究提出了新的需求,即从图片级检索精细化到目标级检索。由于拍照检索中待检索的图片大多是用户在日常生活的真实场景中拍摄的,背景复杂多变且与图片库拍摄环境差异较大,而且通常会包含多个非感兴趣的目标,这给基于内容的目标级图片检索带来较大的技术挑战。针对上述问题,本文围绕交互式目标分割、目标检索展开研究,主要工作如下: 1. 针对以往交互式分割中不能较好地平衡交互复杂度以及交互带来的信息量的问题,提出了一种新的交互式分割算法 LinedCut,该方法利用用户在要提取的目标上画一条线来获取交互信息,交互方式简单便捷。同时,该方法能较好地处理要提取的目标的尺度问题。此外,该方法通过嵌入一个与交互线的距离函数,对前景背景色彩分布差异不是很大的图片,也能提供较好的适应性。 2. 在上述交互式分割算法基础之上,提出了一种基于目标提取的检索算法框架。整个检索过程先通过交互式目标分割提取出待检索的目标前景,然后在这个目标前景而不是整图上提取特征,最后用提取出的特征来进行检索。其中,提取的特征包括颜色特征和基于码本的关键点特征。距离融合方式,在详细讨论了前融合与后融合各自优缺点的基础上,采用后融合的方式。在小规模数据库上的实验结果表明,该方法可有效去除背景干扰,提取复杂形状目标,解决图像中存在着多个目标等问题,从而使得检索效果相较于传统的基于整图的检索有着不错的提升。 3. 基于以上算法,设计并实现了一个基于交互式目标分割的图像检索系统,该系统具有在线图片目标注册、目标检索、查看检索结果的详细信息等功能,系统较好地分离了后台数据处理过程和前台显示样式,具有较高的可扩展性。 总的来说,本文在交互式图像目标分割与检索方法上开展了一些探索研究,将传统的基于内容的图像检索从图片级扩展到目标级,可以为移动互联网时代的拍照检索应用提供一条可行的途径。
Other AbstractWith the spread of mobile terminal with camera and the rapid development of internet technology, we are now coming to ”mobile internet era” and ”reading images era”. Photograph retrieval has become a more and more important entrance for image retrieval. In photograph retrieval, people are more concerned about interested object in one image,such as logo or item on supermarket shelves,than the image itself. This puts forward new demands for the research of content-based-image-retrieval(CBIR) system, i.e., retrieval on object rather than the whole image. On the other hand, recalling the fact that more and more images are shot in peoples’ daily life in photograph retrieval, the background noise in these images are uncontrollable and could be very different from the photograph environment of images in dataset.Besides, many objects that are not interested may exist in the shot image. All these have brought great technical challenges to the CBIR system based on object. To overcome these problems,this thesis focus on the research of interactive image segmentation and object retrieval. Main contents of this thesis can be summarized as follows: 1. We propose a novel method for object segmentation which requires only a single line drawing to identify the object. In contrast to previous interactive image segmentation algorithms, our method balance well between maximizing prior information provided by user and minimizing the effort of user intervention. Besides, our method needs only a single line drawing to identify the object, it is intuitive and simple in interaction mode and can handle the object scale very well. This method embeds a distance function related to the interaction line, and thus can tackle images with indistinct foreground and background colors very well. 2. Based on the above interactive image segmentation algorithm, we propose a new image retrieval framework. This new framework first uses interactive image segmentation algorithm to get object for retrieval. Then it extracts features from this object rather than the whole image. At last, these features are used to retrieval images from the dataset. The features we extracted includes color feature and key-point feature. We combine them using the late fusion method based on the discussion of early fusion and late fusion. Experimentonasmallimagedatasetdemonstratesthatthismethodachievesbetter performance than traditional image retrieval framework,benefiting from that the method can remove backg...
Other Identifier201228014628029
Language中文
Document Type学位论文
Identifierhttp://ir.ia.ac.cn/handle/173211/7742
Collection毕业生_硕士学位论文
Recommended Citation
GB/T 7714
仇浩文. 交互式图像目标分割及检索系统[D]. 中国科学院自动化研究所. 中国科学院大学,2015.
Files in This Item:
File Name/Size DocType Version Access License
CASIA_20122801462802(6015KB) 暂不开放CC BY-NC-SAApplication Full Text
Related Services
Recommend this item
Bookmark
Usage statistics
Export to Endnote
Google Scholar
Similar articles in Google Scholar
[仇浩文]'s Articles
Baidu academic
Similar articles in Baidu academic
[仇浩文]'s Articles
Bing Scholar
Similar articles in Bing Scholar
[仇浩文]'s Articles
Terms of Use
No data!
Social Bookmark/Share
All comments (0)
No comment.
 

Items in the repository are protected by copyright, with all rights reserved, unless otherwise indicated.