Face Recognition Research Under Complex Lighting Environment

Face recognition is a kind of biometric identification technology which is based on facial features.By collecting face images through cameras or other equipment,different algorithms are used to extract facial features which are classified into right types.Face recognition technology has wide application prospects in financial services,public safety and so on.After many years of research and development,face recognition technology has made great process,but there are still many problems to be solved,among them,the influence of complex lighting environment for face recognition is particularly critical.This thesis focuses on the issue of illumination in face recognition research and mainly includes:(1)Research on image preprocessing and face detection.Three image preprocessing algorithms which are based on spatial and frequency domain are studied,including histogram equalization,gamma transform and homomorphic filtering.Then features are compared through experiments with images from database or collection.This thesis uses Adaboost algorithm for face detection.By extracting Haar-like features from face images,face detection classifier is formed after training and cascading strong classifiers.Finally,experiment results have proven its effectiveness.(2)Extract face illumination invariant features.First,establish the illumination model.Then different algorithms which based on Retinex theory and self-quotient image theory are developed for extracting illumination invariant features.Aiming at the problem that there is shadow in the side light when using MSR algorithm,this thesis proposes the spiral path for estimating the image and introduces the local standard deviation transform for further reducing the illumination effect;In order to solve the problem of losing edge details which is caused by Gaussian filter in SQI algorithm,adaptive bilateral filter instead of traditional Gaussian filter is used.The experiment results on Yale B and other database have shown that the improved algorithms can effectively extract the invariant features of face and improve the recognition rate of face recognition.(3)Classifier design and system construction.According to the support vector machine theory,the LibSVM library is selected for training and classifying facial features.The system uses Microsoft Visual Studio 2013 as development tool.Opencv2.4.9 is configured on the MFC project.By integrating image preprocessing,face detection,invariant feature extraction and classification,interface of the MFC project is designed.Finally,experiments are carried out to verify the great robustness of the face recognition system.