| 本文已被：浏览 4949次 下载 7482次 |
码上扫一扫！ |
|
|
| 基于支持向量机的人脸检测训练集增强 |
|
王瑞平1,2,3, 陈 杰4, 山世光1,2, 陈熙霖1,2, 高 文2,5
|
|
1.中国科学院 智能信息处理重点实验室,北京 100190;2.中国科学院 计算技术研究所,北京 100190;3.中国科学院 研究生院,北京 100049;4.哈尔滨工业大学 计算机科学与技术学院,黑龙江 哈尔滨 150001;5.北京大学 信息科学技术学院,北京 100871
|
|
| 摘要: |
| 根据支持向量机(support vector machine,简称SVM)理论,对基于边界的分类算法(geometric approach)而言,类别边界附近的样本通常比其他样本包含有更多的分类信息.基于这一基本思路,以人脸检测问题为例,探讨了对给定训练样本集进行边界增强的问题,并为此而提出了一种基于支持向量机和改进的非线性精简集算法IRS(improved reduced set)的训练集边界样本增强算法,用以扩大训练集并改善其样本分布.其中,所谓IRS算法是指在精简集(reduced set)算法的核函数中嵌入一种新的距离度量——图像欧式距离——来改善其迭代近似性能,IRS可以有效地生成新的、位于类别边界附近的虚拟样本以增强给定训练集.为了验证算法的有效性,采用增强的样本集训练基于AdaBoost的人脸检测器,并在MIT+CMU正面人脸测试库上进行了测试.实验结果表明,通过这种方法能够有效地提高最终分类器的人脸检测性能. |
| 关键词: 人脸检测 支持向量机 精简集算法 AdaBoost |
| DOI： |
| 分类号: |
| 基金项目:Supported by the National Natural Science Foundation of China under Grant Nos.60533030, 60673091, 60772071 (国家自然科学基金); the National High-Tech Research and Development Plan of China under Grant Nos.2006AA01Z122, 2007AA01Z163 (国家高技术研究发展计划(863)); the 100 Talents Program of the Chinese Academy of Sciences (中国科学院"百人计划"); the ISVISION Technology Co. Ltd. (上海银晨智能识别科技有限公司) |
|
| Enhancing Training Set for Face Detection Based on SVM |
|
WANG Rui-Ping,CHEN Jie,SHAN Shi-Guang,CHEN Xi-Lin,GAO Wen
|
| Abstract: |
| According to support vector machines (SVMs), for those geometric approach based classification methods, examples close to the class boundary usually are more informative than others. Taking face detection as an example, this paper addresses the problem of enhancing given training set and presents a nonlinear method to tackle the problem effectively. Based on SVM and improved reduced set algorithm (IRS), the method generates new examples lying close to the face/non-face class boundary to enlarge the original dataset and hence improve its sample distribution. The new IRS algorithm has greatly improved the approximation performance of the original reduced set (RS) method by embedding a new distance metric called image Euclidean distance (IMED) into the kernel function. To verify the generalization capability of the proposed method, the enhanced dataset is used to train an AdaBoost-based face detector and test it on the MIT+CMU frontal face test set. The experimental results show that the original collected database can be enhanced effectively by the proposed method to learn a face detector with improved generalization performance. |
| Key words: face detection SVM (support vector machine) reduced set method AdaBoost |
