基于局部区域稀疏编码的人脸检测

摘要

提出一种基于局部区域稀疏编码的人脸检测方法。首先提取人脸局部区域作为训练样本；然后学习得到一个具有较强判别性的字典，字典中的每个基与人脸各局部区域有明确的对应关系；接着，基于各检测窗口稀疏编码的响应判断人脸某一局部区域是否出现；最后，利用人脸局部区域的检测结果和位置约束进行投票，完成人脸定位。该方法的创新在于将稀疏编码和基于部件模型的思想相结合，实现人脸检测。在Caltech和BioID人脸数据库的实验结果表明：该方法适用于小样本问题，且在遮挡、复杂表情、人脸偏转等情况下具有较好的检测效果。%In this paper, a face detection method based on local region sparse coding is proposed. First, every local face regions are extracted as training sample. Next, a discriminative dictionary whose atoms have explicit relations with local regions is learned. Then the appearance of a particular local region is determined based on the response of its sparse coding for each detection window. Finally, face location is obtained using position constraints and detection results of local regions. The innovation of the proposed method lies in combining sparse coding and part based model for face detection. Experimental results in Caltech and BioID database show that the proposed method is suitable for small sample size problem and has good detection results in case of occlusion, rotation, complex expressions.