Research On Complex Face Detection Based On Deep Learning

Face detection is a computer vision technique that identifies and locates human faces in images or videos.In the field of security,face detection can be used to identify criminal suspects,find missing persons,monitor public places,etc;in the field of human-computer interaction,face detection can be used to identify users and provide personalized services;Detection can be used to implement face morphing,face recognition games,etc.Although conventional face detection has achieved high results,the accuracy of face detection in complex scenes still needs to be improved.The main goal of this paper is to solve the related problems of face detection in low light and small face detection.First of all,aiming at the problem about detect small faces,a detection algorithm based on the TPH detection head is proposed.Improve the detection accuracy of tiny objects through the TPH module,improve the image resolution by introducing sub-pixel convolution to improve the detection accuracy,and replace the original loss function with Focal loss and CIou loss to more accurately reflect the predicted frame and the real frame.The experiment has achieved good results in the authoritative face datasets WIDER FACE.Secondly,for the face detection problem under low light,this paper introduces a self-calibrating lighting framework as a processing tool for low light conditions based on the Retinaface detection algorithm,at the same time introduce the newly proposed efficient backbone network Ghostnet,and proposes a new attention mechanism for the lack of feature extraction capabilities.Experiments show that compared with the original algorithm,both the low-light face dataset Dark face and the regular face dataset Wider face are significantly improved.