Application And Research Of Convolution Neural Networks In The Human Face Detection

In recent years, research in artificial intelligence is increasingly thoroughly, Human face detection has become an independent project. Human face detection technology is also used in photography, surveillance, access control and other fields. Unlike human face recognition, human face detection aims to identify the location of faces in the picture. The convolution neural network is mainly studied and is applied to human face detection, then compared with the traditional AdaBoost algorithm.Research dynamics and methods of human face detection is introduced in the paper, and the basic theories of these methods are analyzed. On this basis, mainly the most classical AdaBoost algorithm is studied in detail and applied to human face detection. In this method, principal component analysis(PCA) is used to reduce the dimension of samples for reducing the amount of computation; then the AdaBoost algorithm is applied to train weak classifiers and a classifier; finally tests are made on images and results and the influencing factors of the algorithm are analyzed.In addition, the artificial neural network and its fundamental theories are introduced in this paper in which human face detection based on convolution neural network is further studied and the principles, network structure and training methods of convolution neural network are introduced, LeNet-5 structure of classical convolution neural network is studied and the number of output unit and feature map of convolution layer is improved. The face samples selected are input to network, the network are trained and tested, then the test results are given and the influencing factors of the algorithm are analyzed.Finally, AdaBoost algorithm and convolution neural network in the detection rate and the running speed were compared, a large number of experiments show that the face detection method based on convolution neural network not only reduces the duplicate detection rate and the error detection rate in the detection rate, in the running speed and extraction features also is better than AdaBoost algorithm.