A face-based authentication system using correlation filters on videos

This work is the result of the beginning of the development of a system for helping visually impaired people to recognize faces and objects. In order to make such a system, the problem of face recognition must be addressed and this is done by employing recognition algorithms, such as CFA - class dependence feature analysis, and a webcam. The objective of this work is to combine, improve, and develop algorithms for face detection and recognition so as to create a software-based system which is able to detect and recognize faces, previously enrolled in a database, obtained from images captured from a webcam. Specifically for the case of CFA, an algorithm for selecting the images that will compose the training set is proposed which reduces the training time by removing redundant images. The training time reduction is about 80%, without impacting identification performance, which is quite significant. Moreover, some techniques that employ the video sequence captured from the webcam in a very simple way are proposed for increasing verification reliability and stability.