A Study Of Face Recognition With Fluctuation Of Light Environment And Implementation On Andriod

With the improvement of computer technology, face recognition technology has made substantial progress in recent decades. However, the existing face recognition systems usually accomplish recognition operations under ideal environment approximating to the laboratory. They have weak generality in dealing with uneven illumination and gesture variation. At the same time, the demand of face recognition systems working well under natural illumination environment increases extremely strong.In the recent years, many companies and research institutions have bring out their face recognition systems. But these products have no generality because they are mostly designed under specific environment. The dominating reason is that they are sensitive to illumination variation and gesture variation. Therefore, the face detection and recognition technology is necessary to further study.1. Generally, a human face recognition system contains three modules, such as image pre-process module, face detection module and face recognition module. In pre-process module, anti-illumination function is the most important component that keeps the subsequent modules work properly. Among different illumination compensation algorithms, Retinex based algorithms have drawn a lot of attention. And on this basis, this paper puts forward an improved adaptive multi-scale Retinex algorithm. At the same time, we validate the effectiveness of the proposed algorithm. The simulation results fully prove that the proposed improved adaptive multi-scale Retinex algorithm is both effective and robust.2. We study the theory of face detection model based on skin and the Adaboost algorithm. The paper uses a multi-stage face detection method. We firstly use skin model for pre-detection, in order to get probable regions of face in the image. Then we accurately detect face in those probable regions with Adaboost algorithm. The experimental results show that the combination of skin model and the Adaboost algorithm makes face detection more rapid and effective.3. We study the basic LBP algorithm, the related model of LBP and the face recognition algorithm based on blocked LBP. And we analyse the influence of the block size on the face recognition. Since the basic LBP algorithm is not perfect and the extracted feature is not complete, a face feature extraction algorithm based on improved LBP is proposed.4. According to the ideas of combining theory with practice, and the knowledge of face recognition, we thoroughly study the Android system architecture and set up a development environment for both application layer and local layer. Then, we design a software development plan based on Andriod platform features, transplant face recognition algorithms library into the Android platform and complete the face image acquisition, face detection and face recognition in the Android mobile phone. Finally, we use huawei cell phone 3c for experiment, and the experimental results show that the system has high accuracy and real-time performance, which realizes the function of real-time face recognition and verifies the correctness of the algorithm. The self-built face database is used, with ten times test under the different environment and gesture. The results show that the recognition rate reaches 83.5%, and the average time is 93.3 ms, which is much smooth with the visual experience.The system achieves the purpose of this project overall, however, there are still lots of parts worthy of improvement. As this article only considers the influence of illumination change on face recognition, and in the actual environment, gesture change is also one of the main factors affecting face recognition accuracy. We can further consider to improve the safety of human face image database in case of the third party criminals.