Liveness Detection Model Training With Masked Face Reconstruction
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Face recognition systems are vulnerable to photo attacks, video attacks, and model attacks, leading to poor recognition accuracy due to the lack of effective live-body detection methods.
Innovation Solution
A live-body detection model training method that includes blocking regions in sample images to create masked images, extracting encoding features using an encoder, restoring features with a decoder, and updating model parameters based on reconstruction loss, combined with contrastive learning to improve feature extraction and recognition accuracy.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If traditional face recognition is used, then the system is simple and fast, but recognition accuracy deteriorates due to photo attacks, video attacks, and model attacks
Solution Approach 1:
The patent segments the face image into multiple regions (e.g., left eye, right eye, nose, mouth) and extracts features from each region separately. This segmentation allows the system to detect live body characteristics in different facial regions, making it more reliable against photo and video attacks while maintaining manageable system complexity through modular processing
Solution Approach 2:
The patent introduces an intermediary live body detection model between the traditional face recognition system and the authentication decision. This intermediary model analyzes facial region features to determine live body status, serving as a mediator that enhances security without completely replacing the original recognition system, thus balancing reliability improvement with complexity control
2Measurement precision
If masked region training is applied, then feature extraction capability is improved, but training complexity and computational cost increase
Solution Approach 1:
The patent applies preliminary action by pre-training the encoder on masked region images before the main training process. During pre-training, random regions of images are masked out, and the encoder learns to extract features from the remaining unmasked regions. This preliminary training enhances feature extraction capability by forcing the model to focus on salient features, while the structured approach to masking and training simplifies the overall process
Solution Approach 2:
The patent uses copying by creating synthetic masked images from existing training images. Instead of requiring complex real-world scenarios, the system copies and modifies existing images by masking regions, generating training data that simulates challenging conditions. This copying approach improves feature extraction capability while keeping training complexity manageable through automated image processing
Data Source
Figure 1
Figure 2~3
Figure 4
AI summary
This application belongs to the field of artificial intelligence technologies, and specifically relates to a live-body detection model training method and apparatus, a medium, an electronic device, and a product. This application may be applied to various scenarios, for example, a cloud technology, artificial intelligence, intelligent traffic, or assisted driving. The training method includes: blocking a part of a region in a sample image to obtain a masked image including a non-masked region, the non-masked region being an unblocked region in the masked image; obtaining an encoding feature of the non-masked region that is obtained by performing feature extraction on the non-masked region by an encoder; obtaining an output image obtained by performing feature restoration on the encoding feature of the non-masked region by a decoder; and updating model parameters of the encoder and the decoder according to a reconstruction loss value between the output image and the sample image. Recognition accuracy of live-body detection may be improved according to embodiments of this application.