CASIA OpenIR  > 毕业生  > 硕士学位论文
Alternative TitleResearches on Human Face Technologies
Thesis Advisor胡包钢
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模式识别与智能系统
Keyword模式识别 人脸特征点 头部姿态 家庭相册 基于内容的图象检索 Pattern Recognition Facial Feature Head Pose Estimation Family Album Content Based Image Retrieval
Abstract人脸技术的研究作为计算机视觉、模式识别和图像处理等研究的重要领域之一,近年来受到越来越多的关注。本文收集了本人在硕士学习期间的关于人脸技术应用问题的一些研究。这些研究包括人脸特征点的检测、头部姿态估计、家庭照片集中的人脸标记等。本文的主要贡献在于:1)提出了一个新的快速鲁棒的人脸特征点定位方法。该方法首先提出了一种快速有效的计算象素概率输出的方法,然后在概率输出的基础上提出了一个人脸特征点定位的方法。该方法的精度和人工标记的结果是可比的,而且对姿态、光照、表情等的变化是鲁棒的。 2)对于头部姿态估计问题,本文提出了一种新的方法,把头部姿态估计的精度从10度左右提高到3度左右。以前的姿态估计的方法主要是用回归方法, 但是由于头部姿态估计的训练样本不容易获取,而且回归算法的逼近能力和泛 化能力在处理高维的图像样本都比较困难,用现有的回归问题的算法往往不能 够得到很好的回归效果。作者提出了用较少的容易获取的样本和基于监督分类 学习的方法来估计头部姿态问题,克服了训练样本的获得和学习结果的泛化能力上的问题,取得了较好的姿态估计结果。 3)为了解决在家庭照片集中的人脸检索问题,我们提出了一个为家庭照片 集作半自动人脸标记的框架。该框架的核心是人脸相似性特征和改进自动人脸 标记结果的算法。除了人脸识别所用的特征,我们采用了图像相似性特征和相 关反馈方法。这些图像特征和检索方法是在基于内容的图像检索的研究中发展 起来的。为了评价我们提出的方法,我们对一个有l,707张照片的家庭集进行 ,模拟标记,实验证明我们提出的方法是有个有效的半自动家庭照片集标记方法。对于计算机视觉技术来说,这是个重要的结果。
Other AbstractFace research and related technology is an important part of computer vision, pattern recognition and image processing. It is attracting more and more attentions in recent years. This thesis collects some of author's researching result and its applications. The author's main contributions are: 1) We proposed a method for facial feature points localization. Our method is extremely fast and the accuracy rate is comparable with hand-labeled results. 2). A new method for head pose estimation is proposed, improving the estimation accuracy from about 10 degrees to about 3 degrees. Previous method on head pose estimation is regression method. However, face images with accurate head poses arc not easy to obtain and due to the large dimensionality of image samples, the approximating and generalizing ability of regression machines are usually incapable for head pose estimation. The approach we proposed uses training samples that arc easily obtained and the classification-based method with good generalization performance to achieve better result than regression methods. 3).In order to build a face retrieval system for family photo album, we proposed a framework of face annotation for family photo album. The core technique of face annotation is the definition of face similarity measurement and the algorithm to improve the result of annotation result. Besides the facial feature for general face recognition, we adopt the relevance feedback approach for similarity-based image retrieval system, which is developed in the research of content-based image retrieval system. To evaluate our approach, we test our algorithm on a photo album containing 1,707 photos. The result proves that our approach is effective for semi-automatic face annotation for family album. Patents has been applied for this technology, and it will be part of new Microsoft' Product. It is an important advance for computer vision research and face technologies.
Other Identifier708
Document Type学位论文
Recommended Citation
GB/T 7714
陈龙斌. 人脸技术应用中若干问题的研究[D]. 中国科学院自动化研究所. 中国科学院研究生院,2003.
Files in This Item:
There are no files associated with this item.
Related Services
Recommend this item
Usage statistics
Export to Endnote
Google Scholar
Similar articles in Google Scholar
[陈龙斌]'s Articles
Baidu academic
Similar articles in Baidu academic
[陈龙斌]'s Articles
Bing Scholar
Similar articles in Bing Scholar
[陈龙斌]'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.