基于级联卷积神经网络的人脸检测算法的研究
Research on Face Detection Algorithm Based on Cascaded Convolutional Neural Networks
Abstract
在人脸识别领域中,人脸检测是其中的一个重要环节,2001年Viola和Jones发明了VJ人脸检测算法,该算法的诞生使人脸检测能够得到实际应用,但对于人脸多角度、多光照、多姿态等问题,VJ人脸检测器表现不佳。DPM算法的引入较好解决了复杂情况下的人脸检测问题,但是该算法存在计算量大,前期人工标记成本大等问题。神经网络特别是卷积神经网络的发展和GPU编程的实现给图像处理、音频处理、视频处理等领域注入了新的活力。卷积神经网络(CNN)在人脸检测领域有着无可比拟的优点,比如自学习特征、较好适应复杂情况下人脸、节省人力成本等。CNN用于人脸检测效果良好,但也存在着解释性较差的缺点。虽然人脸检测算法取得... Face detection is an important step in face recognition. In 2001, Viola and Jones invented the VJ face detector which made it possible in practical use. However, the VJ face detector has poor performance while detecting the face in the wild. The introduction of DPM algorithm solves the problem of face detection in complex cases, but it has large cost in computation and manual marking. Thanks to th...