Study And Implementation Of An Anti-spoofing System Based On Facial Information Analysis

Today's society is an intelligent society. Intelligentization has brought great convenience to people's lives. Face recognition technology is becoming more and more sophisticated, which is a manifestation of intelligentization. The use of face recognition technology for identity authentication occurs in more and more occasions. However, face recognition technology focuses on recognizing the face in the collected image, but can not judge whether the given face image is from the real user or a fake face. With the fast development of Internet, accessing a photo of a user is very easy, so the face recognition system is facing face spoofing attacks. Once succeed, such attacks may bring unpredictable consequences.In order to solve this problem, face anti-spoofing technology came into being. This technique aims to distinguish whether the input image which enters into the face recognition system is from a real person or a fake face image. Combining face detection with face anti-spoofing technology, we can obtain more secure and reliable face recognition performance, which is of great significance in reality.In this paper, after numerous research and literature reading, an anti-spoofing method has been proposed based on dense optical flow feature,and a large number of experiments have been carried out to verify the validity of the proposed method. In addition, a face anti-spoofing detection system based on facial information analysis has been implemented. In particular, the research contents of this paper include the following three parts:1. Feature descriptor of face anti-spoofing detectionFeature descriptor is feature point that used to represent characteristics of image or video. As the basis for face anti-spoofing detection, it has a high degree of uniqueness, in order to ensure accuracy.In this paper, the steps of face anti-spoofing detection include the extraction of feature descriptors for training and classification. In this paper, the sum of displacement of dense optical flow vector between two adjacent frames is chosen as the feature descriptor, which is calculated by polynomial expansion and robust to scale transformation and rotation transformation.Feature comparison experiment is carried out to validate it.2. Matching classifier for face anti-spoofing detectionA classifier is a function or classification model used to predict data types. Different classifiers are suitable for different feature descriptors.Selecting an appropriate classifier can improve the accuracy of the algorithm. In this paper, we choose K-nearest neighbor classifier to classify the extracted feature descriptors on the basis of comparison experiments of several classifiers. The core idea of K-nearest neighbor classifier is that if most of the k nearest neighbors of a sample in a feature space belong to a certain type, the sample belongs to this type and has the characteristics of this type. The K-nearest neighbor method is more suitable than the other methods after classification comparison experiment is carried out.3. Face anti-spoofing detection systemFace anti-spoofing detection system includes three modules: face detection, face anti-spoofing detection and face recognition. Face detection means that for any given image, a certain strategy is used to search whether it contains a face, and if so, return the position, size and attitude of the face.In this paper, we mainly use Adaboost cascade classifier and Haar feature to implement face detection module, which has a high detection success rate. Face recognition technology bases on human facial features. For the input face image or video stream, each face's position, size and positon information of main facial organs are given. And based on these information, further identity features of every face are extracted to compare with the known face to identify the identity of each face. In this paper, we use Uniform LBP feature and K-nearest neighbor classifier to implement face recognition module.In summary, in this paper we mainly study the algorithm of face anti-spoofing detection based on dense optical flow feature, and implement a face anti-spoofing detection system which is of great significance in practical.