Gender detection method based on average face pre-learning

Abstract

The present invention is based on the gender detection method of average face pre-learning, comprises learning step and detection step: Learning step: A, carry out classification to face image database, calculate the average face of all kinds of face images; B, described each average face The data is configured as the output layer of the convolutional neural network, each position of the face in the face data set under the category of each average face is configured as the input layer of the convolutional neural network, and the convolutional neural network is pre-learned; C, with gender The classification layer replaces the output layer of the convolutional neural network, and the convolutional neural network and the gender classification layer are finally learned; the detection step: input the face image of the person to be checked into the convolutional neural network after learning, by The gender classification layer outputs gender. The invention avoids some poor sub-optimal solutions, effectively solves the limitation that the deep neural network cannot obtain a better model at one time through a classical learning method, and improves the accuracy of gender detection through pre-learning steps.

Description

technical field [0001] The invention relates to the field of human face gender recognition of electronic equipment, in particular to a gender detection method based on average face pre-learning. Background technique [0002] In the prior art, gender recognition algorithms are roughly divided into three categories, one is based on voice, the other is based on human gait, and the other is based on facial images. As the current face detection technology is relatively mature, the scheme based on face images becomes simpler and more direct. However, due to factors such as the complex background in the real environment, the difference in the accuracy of the lighting and the camera itself, and the angle of the face, it greatly increases the difficulty of gender recognition, resulting in relatively low accuracy and low stability. [0003] Further detailed analysis shows that the biggest problem of gender recognition technology based on face images is low accuracy and instability. ...

AI Technical Summary

Problems solved by technology

[0011] This application provides a gender detection method based on average face pre-learning. By pre-learning the convolutional layer of the convolutional neural network, it avoids some relatively poor sub-optimal solutions, and effectively solves the current problem of deep neural networks. Some traditional technologies cannot obtain a better model through learning, so as to improve the accuracy of gender detection

Images