The Research On Face Detection Algorithm Based On Mixed Model Of Deep Learning

The face detection technology is one of the important research topics in the field of pattern recognition.In practical applications,the collected face images are often affected by the surrounding environment,resulting in posture changes,occlusion and complex background in face detection,leading to the accuracy and robustness of face detection can not meet the requirements of practical application.Combined with the theory of deep learning,this thesis proposes a face detection algorithm based on the deep learning mixed model.By establishing a deep learning mixed model,using the strong correlation between the features of face to learn the local features and location,so as to reduce the influence of partial occlusion and multi-pose on face detection.The main contents of this thesis are as follows:1.Firstly,combined with the theory of deep learning,this thesis analyzed and studied its three structures and their typical models of deep learning from the aspects of feature extraction,training and convergence time.Secondly,we compared the two methods of Deep Belief Network and Convolutional Neural Network from the perspective of classification error and convergence.The experimental results show that the percentage of classification error using the Convolution Neural Network method is lower than that of the Deep Belief Network method,but in the convergence rate,the convergence rate of the Deep Belief Network is better than that of the Convolutional Neural Network.Therefore,according to the advantages and disadvantages of the two typical models,this thesis constructed a Convolutional Pooling Boltzmann model unit,the pooling layers and convolution layers in the Convolution Neural Network were added to the hidden layer in the Restricted Boltzmann Machine,the purpose was to construct the basic unit of the deep learning mixed model.The improved network can input the original image directly without preprocessing,and its structure is more in line with the topological structure of the image input,also it is more suitable for the training of the image.2.Aiming at the single deep model would appear the problems of low learning efficiency and high false detection rate of face detection in solving the occlusion of face detection,the thesis proposes a mixed model based on deep learning,that is Convolutional Pooling Deep Belief Network,CPDBN.First of all,the constructed Convolution Pool Restricted Boltzmann Machine was used as the basic unit of deepmodel,then built a multi-layer basic unit structure,and used the correlation between the deep structure of the depth model to learn the relationship between the position and characteristics of each feature.When occlusion occurs,according to the local features of the face detected by learning,the hidden feature position is predicted and inferred,face detected by complete face features.Experimental results show that the proposed algorithm accelerates the convergence speed,improves the face detection accuracy under partially occluded,and it has a certain robustness to the change of human face pose.