Design The Face Recognition System Under Complex Illumination

Face recognition is an important research topic of pattern recognition and computer vision, as well as the most promising means of authentication, an economic and accurate face recognition system has various applications. However, there are some difficulties to be solved in face recognition research, such as illumination changes, pose variations, and so on.Illumination change is a long-standing problem in machine vision, and it matters most especially in face recognition. Based on the illumination problem in face recognition, this thesis addresses the face recognition theory and system implementation under complex illumination.According to the general process of face recognition, the research of this topic contains illumination processing of face image, facial feature extraction and classifier design, also completes the automatic face recognition system combined with face detection.During illumination processing, an improved Multi-Scale Retinex algorithm based on Gamma transformation was proposed to remove or reduce the influence of complex illumination such as side light, uniform illumination on the image. And the AdaBoost algorithm which is recognized as the most effective face detection method is applied in the system. In the process of facial feature extraction, the 2-d Gabor wavelet transform was used to get light-independent features. To reduce the calculated amount of recognition process, the module of transformation is chosen as features, and the mean of features that correspond to different frequencies and angles as the global feature of image, PCA and LDA are applied to reduce the dimension of features. In the process of recognition, the classifier for recognition is designed based on SVM. And One-Class SVM is used to determine whether the input belongs to the known faces. Finally, complete the automatic face recognition system by programming.The automatic face recognition system with the abilities of parameters training, face recognition, video capture, etc. is developed in Qt Creator with OpenCV. The experiments on FERET database demonstrate the recognition rate achieves 94%, the system is real-time and reliable, and it can satisfy the requirement of face recognition under complex illumination.