显著性多特征融合人脸识别研究

项目名称： 显著性多特征融合人脸识别研究

项目编号： No.61203261

项目类型： 青年科学基金项目

立项/批准年度： 2013

项目学科： 自动化学科

项目作者： 陈振学

作者单位： 山东大学

项目金额： 24万元

中文摘要： 人脸识别是生物特征识别技术的一种，且具有特定的优势。在人脸识别过程中，不同的人脸特征对人脸识别的贡献存在较大的差异，往往是几个关键性的特征在引导着人脸之间的差别，而多特征融合的人脸识别也是目前发展的趋势之一。因此，人脸特征显著性的评价机制以及多特征的融合方法成为本课题研究的重点。课题组在研究了人类视觉在实际识别目标过程中表现出的序贯性和层次性，将由于多种随机变化引起的特征可靠建模和识别问题转化为目标特征提取、选择和融合识别问题。提出基于显著性的人脸特征提取和特征选择方法，以概率统计为准则来组织和表达人脸的特征，建立人脸特征的显著性层次模型。对于优选出的显著性特征，采用基于最小错误概率的改进DS证据理论决策级融合方式，融合过程中不断更新人脸的置信度函数，充分体现了"越显著的人脸特征对于人脸识别的贡献越大"这个事实。该课题的研究成果，有望对其他类目标识别技术提供一定的参考和借鉴。

中文关键词： 特征提取；显著性；特征选择；特征融合；人脸识别

英文摘要： Face recognition is a kind of biological characteristics recognition technology, and it has special advantage. In face recognition, the different face features are much more different at the contribution of face recognition. In general, several key features lead to differences among many kinds of the human face. Face recognition based on multi-features fusion is also one of the development trend. Therefore, the feature salience evaluation mechanism and multi-features fusion method are the key points of this project. After researching sequentiality and hierarchical represented by human biological vision in the process of identifying target, project group converts features reliable modeling and recognition problem caused by random change to the target features extraction, features selection and fusion recognition. Secondly, project group presents the method based on the salience feature extraction and feature selection. Then, project group organizes and expresses the face features with probability and statistics as criteria, and establishes salience level of face features. Finally, for the selected salience features, research group takes the improved DS evidence theory decision-level fusion which is based on minimum error rate. And, in the fusion process, constantly updates face confidence degree function. This m

英文关键词： Feature Extraction；Salience；Feature Selection；Features Fusion；Face Recognition