Disguise and spoofing detection in Face Recognition

Abstract

In recent years, facial recognition and vulnerability of the face are the most popular. The facial recognition techniques have been used in many access control applications in the world. One of the bottlenecks is that such data can be stolen or duplicated and misused. The main goal of our research is to detect the disguise and spoofing facial recognition to identify the valid user to protect the confidential data of hackers in the applications. The photo or the video of the face of an authorized person are stolen by an unauthorized person and obtain access to services and facilities called spoofing attacks. In addition, the disguise attacks on the face means that the system cannot be access by the original person due to that person have different appearance and variances. So the fake user and the valid user must be identified by the proposed solution. There are two phases used in this proposed solution such as training phase and testing phase. These phases are carried out through five process (Data collection, pre-processing, feature extraction, feature filtering and classification). Combined algorithms are used for achieving the objective of this document such as PM (PCA+MED), LPM (LBP+PCA+MED), HPM(HOG+PCA+MED), SM(SURF+MED) and HM(HARRIS+MED). Images are collected from the online Databases (NUAA contains spoofing and disguise faces, FEI and DFD only contains disguise faces) which is used for training and testing phases. LBP, HOG , SURF and HARRIS algorithms are used for feature extraction. The principal component analysis algorithm (PCA) is applied at the top of the LBP and HOG algorithm to reduce irrelevant features, preserving the most dominant ones. Selecting strongest points algorithm is used for SURF and HARRIS to extract best points. Finally, the MED classifier is applied for each feature vector for classification. The scope of this project is limited to implementing this model only to simulate attacks on photos and disguise attacks with lighting, posture, expression. This can be further extended with the data set of spoofing video images with tracking and the age difference of the face images that will be saved as future jobs.