For example,Бобцов

CONVOLUTIONAL NEURAL NETWORKS FOR FACE ANTI-SPOOFING

Annotation

Subject of Research. We propose the method for detecting an incident at face authentication when imposter tries to disguise himself as a real client. He tries to falsify the client's face by photo or video. Method. The face anti-spoofing method involves two successive steps. Obtaining facial features takes place at the first stage. Classification is performed at the second stage for making a decision if the real person or an imposter is in front of the camera. Facial features are extracted with the use of deep convolutional neural network. Classification is realized by support vector machine. One frame or a group of frames can become the input data for the method. Main Results. The proposed method for anti-spoofing gives the possibility to work with both real faces obtained with low quality and with fake faces displayed on high-resolution displays. It is confirmed by experiments on two available test datasets. Experiments show that the average value of the first and second kind errors on the test data does not exceed 9%, and the accuracy reaches values over 91%. Classification results are comparable with the best results shown when applying the other known techniques for detecting spoofing attacks on the same test bases. Practical Relevance. The proposed method can be applied to improve the quality of face authentication systems, as well as for the development of multimodal biometric systems.

Keywords

Постоянный URL

Articles in current issue

CONTENT © ITMO UNIVERSITY 1993-2024