摘要
提出了一个在复杂背景下检测灰度人脸图像的混合方法 ,它用基于多分辨率的二值马赛克图像和投影分析方法来检测双眼 ,以加快从背景中分离人脸的速度 ;用改进的视网膜抽样模型抽取最能代表人脸特性的低维数据 ;降维后的输入图像用学习矢量量化网络训练特征。实验表明 ,本方法有近 90 %的检测成功率 ,速度较快 ,对部分遮挡的人脸定位也适用。
A new hybrid approach for face localization in a complex background is presented. The method combines an eyepair detector based on binary mosaic images and projection analysis at different resolutions, a modified retinal sampling model and an adaptive feature organization way via learning vector quantization network. The eyepair detector is used to accelerate the segmentation of face candidates, the modified retinal sampling model can derive the most representative data from a high dimension space, and the LVQ network acts as an adaptive feature extractor. Experimental result demonstrates its 90% of successful localization rate and the feasible processing speed, it may be work even in a facial part missing occasion.
出处
《数据采集与处理》
CSCD
2002年第1期37-41,共5页
Journal of Data Acquisition and Processing
基金
国家自然科学基金 (编号 :6 9772 0 2 6 )
广东省自然科学基金 (编号 :970 4 84 )资助项目
