摘要： 为提高传统支持向量机无约束人脸检测算法的检测精度,基于可变形模型思想,将整体与局部特征级联方式结合,提出一种新的人脸快速检测算法。在第一层级中,设计整体人脸稀疏特征,以快速地提供精确的人脸候选区域,在第二层级中进行人脸定位,捕捉无约束条件下人脸拓扑形状,提取关键特征点周围鲁棒性特征,得到判别能力强的分类器验证候选区域。实验结果表明,该算法能流畅运行于VGA视频流中,提高无约束人脸检测精度,有效降低误检率。

关键词: 可变形模型, 支持向量机, 人脸检测, 人脸对齐, 稀疏特征

Abstract: In order to improve the detection precision of Traditional Support Vector Machine(SVM) unconstrained face detection algorithms,based on the thought of deformable parts model which combines global and local feature in a cascaded way,a new face detection method is proposed.In the first layer,sparse global face features are designed to obtain the precision candidate face regions quickly.In the second layer,face alignment is implemented to capture the unconstraint face topology shape.Robust features are extracted from the surrounding of face landmarks to obtain a discriminative classifier which verifies the candidate regions.Experimental results shows that the proposed algorithm runs fast in VGA video,improves the unconstraint face detection accuracy and reduces the false detection rate effectively.

Key words: deformable model, Support Vector Machine(SVM), face detection, face alignment, sparse feature

中图分类号: 

• TP391.41