Face Verification for Mobile Personal Devices

In this thesis, we presented a detailed study of the face verification problem on the mobile device, covering every component of the system. The study includes face detection, registration, normalization, and verification. Furthermore, the information fusion problem is studied to verify face sequences, and to fuse different modalities. Although the work is application-specific, the thesis is not limited to the application, but more extensive. In every step, we have justified the methods we choose both from the theoretical and the practical point of view. In the review part of each chapter, we discussed principles and methodologies on a higher level, for a better understanding of the problems in general. In our solutions, on the other hand, we have taken care of the application requirements, and put much emphasis on the efficiency and simplicity of the methods.

The system has dealt with the security, convenience, and complexity requirements, which are put forward in Chapter 1. As discussed in Chapters 3 and 4, we verify the input pattern in a high-dimensional space, which is in theory sufficiently discriminative and guarantees security. For the user convenience, a low false rejection rate is essential, and this has been taken care of by the illumination normalization and the decision fusion as discussed in Chapter 5, 6, and 7. The time-sequence verification, furthermore, not only improves security, preventing the scenario that the device is taken away after being logged-on, but also increases the user-convenience by lowering the false rejection rate with the fusion rule. Finally, the algorithmic complexity has always been an important concern of our work, and the system has been successfully implemented on the mobile device with limited computational resources.