Face Detection Method Based On Deep Learning

Face detection is a key step in face recognition technology.Face detection has many researches and applications in the field of computer vision.In most research work,the actual scene environment will affect the acquisition of face images,such as light,occlusion,as well as one’s own posture,position,and photographing method,all of which have a certain impact on the accuracy of face detection.Most face detection models have studied this problem but perform poorly.In this thesis,through the research on face detection technology,A face detection model based on deep learning is constructed,and automatic feature extraction is performed on images to extract features that are more conducive to face detection,so as to solve the influence of external environment and own factors on face detection.This thesis first introduces the face detection process,including candidate region selection,feature extraction,classifier classification and other steps.Carry out research work according to this process to more accurately select candidate frames that may contain faces from the image;then use deep learning technology to extract features from the images in the candidate frames,and extract features that are more representative of faces;finally use A classifier is used to classify the extracted features and determine whether there is a face in the image.The main content of the thesis are as follows:(1)Aiming at the interference of the external environment on the face detection algorithm,this thesis proposes a convolutional neural network face detection algorithm based on skin color distinction.First,affine transformation is used for face alignment,and then the reference white algorithm is used to perform light compensation on the image;the skin color detection based on YCr Cb space is used to extract face features,and finally the convolutional neural network combined with Adaboost algorithm is used to train the generated classifier.In the case of occlusion and multi-pose,the algorithm in this chapter can maintain high accuracy and has strong robustness.The detection accuracy reached98.6%(2)The traditional face detection algorithm based on image pyramid mechanism often loses the face detail information from the extracted high-level semantic features,while the low-level features have the face detail information.Further improve the expression ability of the features extracted by the face detection algorithm to the face,and solve the impact of the complexity of the face on the performance of the face detection algorithm.This thesis proposes a multi-layer fusion algorithm based on an improved attention mechanism,so as to obtain high-level feature vectors with more low-level details and high-level abstractions,so as to improve the richness and representativeness of feature vectors,and avoid the complexity and The external environment interferes with the face detection algorithm.The algorithm combines deep learning theory with the idea of partial model-based face detection.(3)The Tensor Flow deep learning framework is used to build a face detection platform,and the face detection model proposed in this thesis is deployed on this platform to further verify the effectiveness and practicability of the algorithm model proposed in this thesis.