Research On Gender Classification With Deep Learning Methods

In recent years, the information security and privacy protection have become a vital issue in the society. With the rapid development of computer science and technology, these fields, including information, digital and intelligence, have become the new development direction. Face is one of the most significant biological features of human beings, and face image contains a great deal of information, including identity, gender, age, ethnicity, and expression, which plays an exceedingly critical role in the gender classification task.This work explored the gender classification method through human face images based on deep learning model and carried out a great deal of experiments on FERET and CAS-PEAL-R1 datasets. Then, we have proposed the improve methods for stacked-autoencoders model and convolutional neural networks model respectively. The experiments show that our methods are able to improve the classification accuracy for gender classification task. In addition, the author has also proposed the improved weights assignment method for ensemble model in order to improve the accuracy and stability of output from the classifier, which is able to decrease the misclassification rates for this task and acquires higher accuracy than basic model averagely.At last, we have also developed a real-time gender recognition system with the ensemble model and applied it to a recommendation scenario in shopping mall, where the input images captured from the camera and marked the classification result through model forecasting. This system enables to meet real-time response and high accuracy requirements.