摘要
针对传统的人脸识别方法在处理多姿态变化人脸问题时耗时过长及成本过高的问题,提出了一种基于尺度不变特征融合(FSIF)的双阶段分类方法。首先执行均值漂移线性判别分析找到5个类似于查询人脸的最佳候选类;然后利用尺度不变特征融合提取出候选人脸及查询人脸的融合特征描述符,并进行特征匹配得到目标人脸;最后,根据特征描述符的匹配关键点数目完成人脸的识别。实验结果表明,所提方法解决了由姿态变化引起的大的脸部变异,降低FSIF人脸识别的计算复杂度,并在不降低识别性能的前提下大大节约了成本,相比于几种较为先进的3D变换方法,所提方法取得了更好的识别效果,有望应用于实时安全系统。
For the issue that traditional methods take too long time and high costs in processing face reeognition with multiple poses, double stages clas- sifications ba^d on fusion of seale invariant features （FSIF） is proposed. Firstly, shifting mean linear discriminant analysis is used to find five oi^timal candidate classes similar to query faces. Then, Fusion of scale invariant features is used to extract fusion feature descriptors of candidate faces and query face and feature matching is dane so as to getting objective face. Finally, matching point number of the feature descriptors is referred to finish face recog- nition. Experimental results show that proposed method can solve the large face variability due to pose variations, decrease the computational time of FSIF-based face recognition, and save costs clearly without decreasing the recognition performance. Proposed method will be hopefully applied into safety real-time operating system as with better recognition efficieney than sveral advanced 3D transform approaches.
出处
《电视技术》
北大核心
2014年第9期223-227,共5页
Video Engineering
基金
河南省教育厅科学技术研究重点项目(13A520504)
关键词
人脸识别
多姿态变化
尺度不变特征融合
双阶段分类
均值漂移线性判别分析
face recognition
multi-poses variations
fusion of scale invariant features
double stage classification
shifting mean linear discriminant analysis
