Research On Large Scale Face Recognition In The Mobile Environment

After nearly 50 years development,Face recognition technology has made rapid progress and significant advancements.It has been widely applied in human computer interaction,public security,e-commerce,family entertainment,etc.However,there are several problems that need to be solved under imperfect conditions.With the popularity of mobile communication network and WIFI,people rely more and more on mobile devices.Therefore face recognition applications under mobile environment have received more and more attention.This paper discusses the effect of illumination variations and face recognition in large-scale database,and proposes several new algorithms.Based on those work,a robust face recognition system under mobile environment is designed and implemented.The main research work and contributions of the thesis are as follows:(1)Analysis on the illumination effect in wavelet domain and a wavelet-based illumination preprocessing algorithm is proposed.Here,by one-level discrete wavelet transform,a given face image is first decomposed into low frequency and high frequency components,respectively.Then the low frequency is processed by SQI to eliminate the effect of illumination variations and the high frequency remains unchanged.Finally the preprocessed image is obtained through the inverse discrete wavelet transform.The proposed method can greatly reduce the computational complexity and effectively reduce the effect of illumination variations on face recognition.Therefore,it is suitable for face recognition application under mobile environment.(2)Research on large-scale face recognition and a cluster based fast face recognition algorithm is proposed.In the training process,we partition the large-scale data into k clusters using the k-means clustering algorithm.Then,all the features in each cluster are formed into a kd-tree.Given an input image to recognize,we calculate the Euclidean distances to determine which cluster it belongs to.We then employ the kd-tree based nearest neighborhood search in the matched cluster,without matching every image feature in the large-scale database.Experiments on CAS-PEAL and self-collected database show the proposed method significantly accelerates face recognition process without sacrificing recognition rate.(3)A face recognition system based on Android platform has been designed and implemented.The system is based on the background server,so its storage capacity and computing capability is stronger.It realizes functions including face detection,face recognition,security password authentication,Screen fill light,etc.