Abnormal face recognition living body detection method and system based on MCCAE network and Dep SVDD network
A technology of face recognition and liveness detection, which is applied in the field of abnormal face recognition and liveness detection, can solve the problems of waste of manpower and material resources, impossible collection of attack types, 3D mask spoofing attacks, etc., to save manpower and material resources, improve generalization performance, The effect of strong generalization performance
- Summary
- Abstract
- Description
- Claims
- Application Information
AI Technical Summary
Problems solved by technology
Method used
Image
Examples
Embodiment approach
[0049] Such as figure 1 As shown, an abnormal face recognition living body detection method based on MCCAE network, including.
[0050] S1. Obtain the same group of multi-channel face data that has undergone image alignment from the data set Unified preprocessing of multi-channel data Among them, the multi-channel data set includes at least the following channels: grayscale image, depth image, near-infrared image, and thermal infrared image.
[0051] S2. Input the processed data into the trained deep self-encoder network MCCAE, use the deep self-encoder network MCCAE to obtain the latent layer features of the real face, and perform image reconstruction for the latent layer features, and the reconstructed The image is output; according to the input image and the output image, the attack score of image reconstruction is calculated;
[0052] S3. Identify whether the input image data is normal face data according to the attack score.
[0053] Specifically, S201, constructing...
specific Embodiment
[0103] This experiment is performed on the WMCA dataset. The dataset contains 1679 real faces and spoofing attack videos of 72 people. There are 347 and 1332 real face data and spoofing attack data respectively. The multi-channel data set contains 4 channels, which are grayscale image, depth image, near-infrared image, and thermal infrared image. Each video sampled 50 frames, and the image resolution was 128x128 pixels. Multi-channel data was registered and image normalization was performed. The invention divides the WMCA data set into a training set (randomly selecting 90% normal faces in the WMCA data set) and a test set (remaining 10% normal face images and spoofing attack images).
[0104] In order to prove the effectiveness and generalization performance of the method, the ROC curve is selected to represent the experimental standard, and by changing the size of the feature dimension L of the latent layer, the optimal result L=256 is finally selected for display.
[010...
PUM
Login to View More Abstract
Description
Claims
Application Information
Login to View More 


