Research And Implementation Of Gender Recognition And Age Estimation By Face Based On Deep Learning

With the improvement of the processing power of massive data by computers,face recognition technology has been rapidly improved,and it has also become a hot topic for many researchers.The gender and age of the face are undoubtedly one of the important information for human identification.And face gender recognition and age estimation have broad application prospects in many fields such as intelligent business,population census,and population aging analysis.Therefore,this paper uses the deep learning method to conduct gender recognition and age range estimation on the Adience datasets uploaded by volunteers using smartphones about gender and age.The main works in this paper are:(1)Data preprocessing:The face region is detected by the MTCNN face detection method,and cropped by the program,At the same time,the blurred image is enhanced by the color histogram equalization.After data preprocessed,the complex background is removed,and the redundant features are reduced to improve the training accuracy,remaining the effective features of the image.(2)Construction of recognition model:The gender and age group identification are trained by convolutional neural network(CNN)and residual network respectively.And well trained model is tested.The results showed that gender CNN model has an accuracy rate of 88.70%upon gender recognition;for gender identification,the gender residual network model has an accuracy rate of 92.73%;for age group,the age CNN model has an accuracy rate of 71.34%,and the age residual network model has an accuracy rate of 78.29%for age group.(3)System design and implementation:System software based on gender recognition and age estimation is designed on Python platform.The system can realize real-time image acquisition by the camera and manual input of the picture,then passing the obtained picture to the MTCNN module,detecting the face through the MTCNN module,and then transmitting the detected face area to the trained gender identification and age range estimation,which it is identified.Finally the interface displays the gender and age segments of the face image for output.The experimental results show that the residual network has a 4.03%improvement in accuracy compared with CNN in terms of gender identification.The CNN accuracy rate has increased by 6.95%.It is thus proved that the residual network is effective for improving the accuracy of the model compared with CNN.