摘要
针对人脸检测中的部分遮挡和多姿态问题,提出一种基于深度学习的人脸检测算法。首先利用部件检测器检测人脸局部区域,然后将人脸局部区域检测结果输入到深度模型中,学习各局部区域之间的相关性,完成人脸检测。该算法将深度学习理论与基于部分模型的思想相结合,实现人脸检测。实验结果表明,该算法在遮挡、多姿态等情况下具有良好的鲁棒性。
To solve the problems of partial occlusion and multi-pose in face detection,we proposed a deep learning-based face detection algorithm. First we utilised face component detector to detect local areas of face,then inputted the detecting results to deep model. By doing so,the algorithm learns the correlation between each local area to complete face detection. This algorithm combines deep learning theory with part model-based idea to implement the face detection. Experimental results showed that the proposed algorithm had good robustness under the conditions of occlusion and multi-pose.
出处
《计算机应用与软件》
CSCD
2015年第12期123-127,共5页
Computer Applications and Software
基金
国家自然科学基金(61263019)
甘肃省教育厅科研基金项目(2014A-125)
甘肃省青年科技基金计划(1506RJYA111)
关键词
人脸检测
深度学习
部分模型
检测精度
虚警率
Face detection
Deep learning
Part model
Detection precision
False alarm rate
