遮挡条件下实时人脸检测

Real-time face detection under occlusion condition

针对遮挡条件下人脸检测召回率不高和训练样本不足的问题,提出了一种数据增强的方法。计算训练样本的平均脸图像,对样本进行手动遮挡处理;使用平均脸策略和图像分割理论对遮挡区域进行分割和处理,获得类似于自然环境下的遮挡样本;使用处理后的样本训练Ada Boost人脸检测器。实验结果表明:通过该方法训练的人脸检测器能够有效地提高遮挡条件下的人脸召回率,同时可以达到实时检测的效果。

Aiming at problem of low recall rate of face detection and insufficient training samples under occlusion condition,a method for data enhancement is proposed. Average face image of the training sample is calculated and manually occluded. Average face strategy and image segmentation theory are used to segment and process the occlusion area to obtain occlusion samples similar to natural environment. Processed samples are used to train the AdaBoost face detector. The experimental results show that the trained face detector can effectively improve the face recall rate under occlusion condition and achieve effect of real-time detection.