In-Depth and Comprehensive Biometrics Research Possibilities on the Face

Nowadays, biometrics has become the most popular recognition system for protecting privacy and important information. However, the recognition performance is sensitive to the changes of shooting angle, lighting conditions, as well as the environments. To deal with these issues, a three-year research project is conducted as follows. 1) High efficiency pose invariant face recognition system 2) Human eye based identification system3) A new ear recognition system with geometric correction and coordinate correspondence In the progress of the first-year project, we use the generative adversarial network to form a pose-invariant recognition system. Through generative adversarial network method, we can solve lacking required information problem and unnatural images that morph method produce, getting a more robust system even in large angle profile face and higher accuracy in profile face recognition. In second year’s project, a high accuracy human eye system is proposed and it leverages the features of texture, geometric, and the property of double eye-lid for high efficiency. This system is proposed for a high tolerance in terms of normal eyes, blinking eyes, or even closed eyes and achieve a stable recognition rate. In the third-year project, it mainly offers a capability of identity recognition from any view-angle. In addition, this project considers a state-of-the-art accelerated-KAZE (A-KAZE) to extract the key points of the ears for feature stability. These points will then be transformed for rotation correction. Finally, the ROC will be used as the major measure for performance evaluation. This project propose a brand new identification system, which can identify people even when the face features are obscured, and it’s more flexible and robust comparing with the state-of-the-art identification system.

Project ID:PB10907-2864
External Project ID:MOST109-2221-E182-048