CASIA OpenIR  > 毕业生  > 硕士学位论文
基于主动外观模型的人脸图像合成
Alternative TitleFacial Image Composition based on Active Appearance Model
王红侠
Subtype工学硕士
Thesis Advisor潘春洪
2008-05-25
Degree Grantor中国科学院研究生院
Place of Conferral中国科学院自动化研究所
Degree Discipline计算机应用技术
Keyword图像合成 主动外观模型 泊松方程 薄板样条 抠图 Image Composition Active Appearance Model Poisson Equation Thin-plate Splines Image Matting
Abstract在计算机视觉研究领域中,人脸图像分析一直是一个很受关注的研究课题。人脸图像处理包括很多方面,比如:人脸识别、人脸检测、人脸表情处理以及人脸编辑等,而且很多领域的研究结果已经应用到了实际生活中。 本文主要是讨论人脸编辑领域中的人脸合成技术。人脸合成技术在日常生活中随处可见,目前网络上流行的'移花接木'技术就是利用PhotoShop软件来合成新的人脸图像。 本文首先对图像合成的相关技术做了回顾,比如抠图,泊松图像编辑等。研究发现大部分的合成技术都需要用户复杂的手工操作,而且合成图像质量的好坏很大程度上依赖于用户操作的精确性。为了克服图像合成中的这个普遍问题,我们在传统图像合成技术的基础上,提出了一种新的处理人脸图像合成方法---基于主动外观模型的人脸合成系统。 该系统是把人脸模型与传统的图像合成技术相结合,提供了一个用户操作方便的人脸合成系统。用户输入两幅人脸图像(源图像和目标图像)以及要编辑区域的语义信息,比如'嘴巴'、'眼睛'者'脸'等,系统将自动地对源图像的编辑区域进行检测、匹配、最后无缝地融合到目标图像中。与传统的人脸合成技术的不同点在于,我们利用学习出的人脸模型对每一幅人脸图像进行搜索和分析,并自动地生成编辑区域的边界信息。大量的实验结果证实了该人脸图像合成系统的有效性。 系统主要由三个模块组成:模型匹配、特征合成和遮挡处理。我们利用主动外观模型作为系统模型对人脸进行搜索分析。本文首先对主动外观模型的训练和搜索进行详细的讲解,接着讨论并分析各种图像匹配算法的优缺点,最终总结出适合处理人脸图像的匹配算法---薄板样条算法,然后我们介绍传统的基于梯度场的图像编辑算法,最后着重介绍自动人脸合成系统的工作原理并展示了大量的实验结果。另外,系统还采用抠图方法解决了由于目标图像中部分人脸被物体遮挡而造成的失真问题。 最后,本文对上述工作进行总结并对其可能的扩展做了展望。
Other AbstractFacial image processing has already been a popular issue in the field of computer vision. Most of the research results are used in the daily life,such as 'image recognition', 'image detection', 'the processing of facial expression' and 'image editing',etc. In this paper, we discuss the sub-problem of image editing---facial image composition. In the internet, it is very popular to use the software of Photoshop to composite new facial images from two or more facial images. At first, this thesis gives a brief retrospect to the related work of image composition. Most of the image composition techniques need tedious user interactions and the results of composition mostly depend on the accuracy of the user interactions. In order to overcome the problem, we propose a new method to deal with the facial image composition based on the traditional image composition---\Facial Image Composition based on Active Appearance Model. This method is an application combined the facial model with the algorithms of image composition. The manual interaction in the system is very simple. The user inputs two facial images(the source image and the target image) and select the semantic information of the editing region, such as the simple words 'face', 'mouth', 'eyes', etc, then our system can automatically search, analyze the input images and seamlessly composite the two images together. The difference between our system and the traditional methods is that we use the facial model to search each input image, and obtain the boundary location of the editing region automatically. The visually satisfactory results demonstrate the effectiveness of our facial image composition system. The system consists of three modules: model fitting, component composition and occlusion processing. We use the Active Appearance Model as the facial model to search each input image. At first, we make great detail of the training and searching of the facial model; then we discuss and analyze the advantages and disadvantages of the image alignment techniques. As a result, we select the Thin-Plate Slpines for the facial image alignment. We use the Poisson equation for the seamless composition. In the last part, we introduce the framework of our system and illustrate some results of facial composition. At last, we make a conclusion of our work and then discuss some possible extensions of our research.
shelfnumXWLW1181
Other Identifier200528014628051
Language中文
Document Type学位论文
Identifierhttp://ir.ia.ac.cn/handle/173211/7429
Collection毕业生_硕士学位论文
Recommended Citation
GB/T 7714
王红侠. 基于主动外观模型的人脸图像合成[D]. 中国科学院自动化研究所. 中国科学院研究生院,2008.
Files in This Item:
File Name/Size DocType Version Access License
CASIA_20052801462805(8303KB) 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
[王红侠]'s Articles
Baidu academic
Similar articles in Baidu academic
[王红侠]'s Articles
Bing Scholar
Similar articles in Bing Scholar
[王红侠]'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.