A Method For Single Sample Per Person Face Recognition With Non-ideal Conditions

With the development of technology and the increase of practical demands, face recognition(FR) technology has become more and more popular in many application fields, such as information security, law enforcement and surveillance, smart cards, access control, and so on. With the increased attention from researchers, many methods have been proposed in the literature. As the reality application environment is very complex, there are still many challenges need to be faced in FR field.One of the most challenges is single sample per person(SSPP) problem, especially for SSPP with non-ideal conditions. For some FR applications, such as law enhancement,e-passport and ID card, there is usually only a single sample per person recorded in the systems. Meanwhile, the probe samples usually contain large appearance variations caused by illumination, expression, age, pose, and so on. Furthermore, the non-ideal information in probe samples can not be completely avoided in the process of collecting face images.Thus, SSPP problem with non-ideal conditions is of great importance for real-world face recognition systems.The author's major contributions are outlined as follows:? In order to solve SSPP scenario with non-ideal conditions, we propose Variational Feature Representation-based Classification(VFRC) method. General learning and collaborative representation are used in VFRC to get the variational feature with respect to the ideal gallery samples. Thus, a corresponding normal feature, which reserve the identity information of the probe sample, is obtained by using the probe sample and the variational feature. A combination of the normal feature and the probe sample is used in the recognition model, which makes VFRC more effective and robust for SSPP scenario.? In order to verify the effectiveness of VFRC, we implement VFRC and several effective FR methods, which can be applied to SSPP scenario, and several FR methods designed to solve SSPP on three standard face databases(AR database, Extended Yale B database and CMU-PIE database) and one challenging database(LFW database). The numeral results of plenty experiments show that VFRC can achieve high recognition rates and keep strong robustness.