| 郭耸,顾国昌,蔡则苏,刘海波,沈晶.基于组件距离匹配度函数检测部分遮挡的人脸方法[J].高技术通讯(中文),2012,22(4):382~387 |
| 基于组件距离匹配度函数检测部分遮挡的人脸方法 |
| A method for detection of partially occluded faces based on component distance matching degree function |
| 修订日期：2010-10-26 |
| DOI： |
| 中文关键词: 人脸检测， 基于组件， 部分遮挡， AdaBoostSVM， 组件距离匹配度 |
| 英文关键词: face detection, component based, partial occlusions, AdaboostSVM, component distance matching degre |
| 基金项目:863计划(2006AA04Z259)，中央高校基本科研业务费专项资金(HEUCF100606，HEUCF100604，HEUCFZ1010)和国家教育部博士点专项基金(20092304120013)资助项目 |
| 作者 | 单位 | | 郭耸 | 哈尔滨工程大学计算机科学与技术学院 | | 顾国昌 | 哈尔滨工程大学计算机科学与技术学院 | | 蔡则苏 | 哈尔滨工业大学计算机科学与技术学院 | | 刘海波 | 哈尔滨工程大学计算机科学与技术学院 | | 沈晶 | 哈尔滨工程大学计算机科学与技术学院 |
|
| 摘要点击次数: 3031 |
| 全文下载次数: 2241 |
| 中文摘要: |
| 为了较好地解决被部分遮挡的人脸的检测问题，提出一种基于组件距离匹配度函数的人脸检测方法。该算法首先采用基于支持向量机的Adaboost算法即AdaboostSVM算法对输入图像进行人脸各组件的检测，然后根据组件间距离的均值和方差构建组件距离匹配度函数，并提出基于组件距离匹配度函数的组件整合验证算法，在此基础上对被遮挡的组件的位置进行估计，进而实现部分遮挡的人脸的检测与定位。实验结果表明，该方法能够提高部分遮挡的人脸的检测性能与速度，能够在复杂背景下对被部分遮挡的人脸进行较为准确的定位与检测。 |
| 英文摘要: |
| To solve the problem of detecting the faces with partial occlusions better, a face detection method based on the component distance matching degree function was proposed. The method was carried out in the study as below: Firstly, the AdaboostSVM algorithm, i.e. the Adaboost algorithm based on a support vector machine (SVM), was used to detect the components of the faces included in the input image. Then, the component distance matching degree function was given, which was determined by the mean and variance of the components’ distance together. The component integration verification algorithm based on the function of the component distance matching degree was proposed, and the locations of the occluded components were estimated. Finally, the faces with partial occlusions were detected and located. The experimental results show that the proposed algorithm can enhance the performance and the speed of detection of partially occluded faces, and can detect the faces with partial occlusions exactly in a complex background. |
|
查看全文
查看/发表评论 下载PDF阅读器 |
| 关闭 |
|
|
|
