Convolutional neural network generation method, age identification method and related device

Abstract

The invention discloses a convolutional neural network generation method, an age identification method, a related device and calculating equipment which are used for carrying out age identification on a face in an image. The convolutional neural network generation method comprises the following steps of training a first convolutional neural network, wherein the first convolutional neural network includes a plurality of convolution groups, a plurality of full connection layers and a first classifier which are successively connected; carrying out corresponding replacement on parts of full connection layers and the first classifier in the trained first convolutional neural network so as to generate and train a second convolutional neural network; adding a new full connection layer and a classifier to the trained second convolutional neural network so as to generate and train a third convolutional neural network; and adding a new full connection layer and a classifier to the trained third convolutional neural network so as to generate and train a fourth convolutional neural network, wherein before the above each convolutional neural network is trained, face image information used for training can be preprocessed.

Description

technical field [0001] The present invention relates to the technical field of image processing, in particular to a method for generating a convolutional neural network for age recognition of faces in images, an age recognition method, related devices and computing equipment. Background technique [0002] As one of the important biological characteristics, the face image contains a lot of information, such as age, gender, race, etc. With the further development of face image research in image processing technology, especially in face age recognition, the face age recognition method based on convolutional neural network (CNN: Convolutional Neural Network) has gradually developed. It plays an important role in many real-life scenarios. In "Computer Science" in 2014, Karen Simonyan and Andrew Zisserman published a paper called "Very Deep Convolutional Networks for Large-Scale Image Recognition", proposing a deeper convolutional neural network model, VGG (Visual Geometry Group...

AI Technical Summary

Problems solved by technology

[0003] However, for face images of different age groups, there is a large gap in the range of change, especially for infants aged 0-5, the change is very large, but for adults aged 25-30, the change is very small, which means A phenomenon that has a greater impact on age recognition

In the existing face age recognition method, the age distribution value is used instead of the true age value, which alleviates the above problems to a certain extent, but the obtained result is in a certain age range, which is not accurate enough. Once the critical value of the range is wrong, Then the estimated value of age will fall into the adjacent interval, resulting in a large error

Moreover, due to the large numerical span of age, even if the above-mentioned VGG model is used for age identification, it is necessary to adopt multiple models according to the distribution of age groups, which is relatively complicated and has low versatility.

Images