Face Detection Based On Deep Learning And Gender And Age Identification

Face detection is a primary goal in computer vision,and also a research direction that is extremely relevant to people's life,because the face is the most important appearance feature of people.It can make people's life more convenient,safe and automatic to detect,recognize and analyze the facial feature information intelligently by computer,and make use of the information for further processing.Face detection technology has made a breakthrough in recent years.At first,face detection could only detect the position of the face in the picture without background.Now,some face detection technologies can accurately detect faces from various angles in natural scenes.You can tell if two faces are the same person;It can analyze people's age,gender and facial expressions based on their faces.In this paper,we study the face detection of age and gender identification is based on neural network,on the basis of the research ideas are as follows: public,with face with age and gender tags database is not much,high quality data for the training of the model is very important,this article choose the database IMDB-WIKI,the gender and age identification is more authoritative database.For the original photo data,we need to select the data that contains face,the tag is valid and each photo only contains one face.The concatenated convolutional neural network(MTCNN)is used to detect the exact position of the face and then align it.And the face picture of 171*171 is cropped for the input of the following model recognition.Inception-Resnet-V1 is used to construct the network architecture for predicting gender and age.In the final full connective layer,we set gender as binary and age as 101.In the construction process of the whole network,it is necessary to compare and select data enhancement and optimization data quality,L2 weight regularization,cross entropy loss function,ladder exponential decay method of learning rate and Adam optimizer to optimize the model.As shown above,the model we built has achieved high accuracy in the classification of face gender and age.The overall trend of the loss curve declines consistently from 3.25 at the beginning to 0.78.The accuracy of gender verification data set also reached 99.99% in the end.Because age is easily affected by external environment,the output result of main calculation plus the accuracy on the interval of plus or minus 1,its accuracy also increases steadily in the training process,and finally reaches 99.37%.Based on the above results,our model has achieved good results in predicting the gender and age of human faces.