Face age estimation method capable of carrying out distributed learning on basis of convolutional neural network

Abstract

The invention discloses a face age estimation method capable of carrying out distributed learning on the basis of a convolutional neural network. The overall process of the invention includes the following steps that: data are extracted so as to form an age data set; the age data set is divided into a training set and a verification set; the last fully-connected layer of a deep neural network is followed by a softmax layer; age estimation network training is performed; softmax loss and mean-variance loss are used together as supervised signals so as to adjust the network; the trained network model is evaluated, so that a network model with optimal performance can be selected; and age prediction is carried out on the basis of the obtained model. According to the method of the invention, thenew supervised signals, namely the mean-variance loss, are adopted, and therefore, the nature of correlation between ages is fully utilized, operation such as the manual introduction of variance is avoided, and any manual intervention except for preprocessing is not required.

Description

technical field [0001] The present invention relates to an estimation method, in particular to a face age estimation method based on convolutional neural network for distribution learning. Background technique [0002] At present, the technology of age estimation through face is mainly divided into two categories. One is to use traditional machine learning methods to extract features, and then design and optimize the objective function for the extracted features, so as to obtain the age to be estimated. The current traditional machine learning method is more about exploring the influence of different features on age estimation, so as to achieve better feature extraction; another method is to use deep learning technology to achieve feature extraction and target end-to-end The entire task of function optimization. In terms of deep learning, different network structures are used for training, and its essence is still to obtain different features through different network struc...

AI Technical Summary

Problems solved by technology

By using the order of age, the idea based on ranking (sequence) is proposed to solve the regression problem into a classification problem, but this method ignores the order of age and the relationship between ages, making it difficult for features to contain these Very useful information for estimating age

There is also the use of the nature of the age distribution to optimize the distribution by assuming a variance to generate a distribution. This method needs to assume a variance to carry out follow-up work, resulting in human intervention, so the age distribution is still not ideal.

Images