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 Ada Boost 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.