A method, device and equipment for face detection based on migration learning

Abstract

The invention discloses a method, device, device and computer-readable storage medium for face detection based on transfer learning, including: performing face images in the collected target data set according to the size of the face images in the source data set Normalization processing; directly migrate the migration layer of the neural network of the source data set, fine-tune the non-transfer layer of the neural network of the source data set, and obtain the grid structure of the convolutional neural network of the target data set; The convolutional neural network after the grid structure of the data set is determined is trained to obtain the target grid parameters of the convolutional neural network of the target data set; The convolutional neural network of the target data is used to identify the real face images in the target data set. The method, device, equipment and computer-readable storage medium provided by the present invention can quickly realize the redesign and training of the convolutional neural network of the data set.

Description

technical field [0001] The present invention relates to the technical field of face recognition, and in particular, to a method, apparatus, device and computer-readable storage medium for face detection based on transfer learning. Background technique [0002] In a modern information society, with rapid technological updates and iterations, it is more and more important to protect personal privacy and information property security. Traditional authentication is through keys, signatures, seals, ID cards, passwords, etc. These verification methods need to be memorized and carried, which are not only easy to forget or lose, but also easy to crack and have a low safety factor. The use of face live detection equipment based on pattern recognition is convenient and fast, with high recognition rate and strong security. However, the existing face live detection algorithms are still poor in distinguishing between real and fake faces, and the training sample set is relatively single...

AI Technical Summary

Problems solved by technology

However, the existing face liveness detection algorithm is still poor in distinguishing between real and fake faces, and the training sample set is relatively single, that is, the existing training sample set generally comes from the same scene, the images are collected from the same device, and the photo forgery mode Similarly, once a new forged image appears, the recognition effect of the new data set will inevitably become worse when dealing with different scenes and the photo format changes. The detection effect of the trained model

[0003] When training a neural network in the existing technology, the data set used for training is generally a certain public data set, which leads to the fact that the model trained under a certain data set cannot get good results on another data set

At this time, for different data sets (new forged face images), the existing scheme needs to redesign and train the neural network for the new data set, that is to say, the number of neural network layers, the number of neurons, and the parameters all need Repeated training and modification many times to find the optimal network structure and optimal parameters, the convergence speed is slow, and the required training time is long

Images