Research On Fast Retrieval Algorithm Of Face Recognition And Development Of Clearance System

At present,the staffs of the railway station are still using manual inspection method in passengers clearance process.This method is time-consuming and laborious,which directly affects the efficiency of passenger clearance.Face recognition technology is developing rapidly,which has the advantages of non-contact and non-mandatory.Face recognition technology has been widely used in security,customs clearance and other fields.Therefore,the automatic face recognition technology can be applied to train station clearance scene to improve the efficiency of customs clearance.With the development of storage technology,the computer can store the image size is more and more big.We are faced with large-scale retrieval of face data,and improving the efficiency of face retrieval becomes very important.This thesis focuses on the research of face retrieval methods,applying the face recognition technology to the train station clearance scene,designing and implementing intelligent face recognition clearance system.In the aspect of face feature extraction,We analyze the existing relevant methods in this thesis.Then,We choose the local feature extraction algorithm LBP with better performance in face recognition.Further,compared to the whole image LBP features,the LBP features based on block can reflect the details of the human face texture features better,and it has strong robustness to illumination changes.Therefore,I make sub-block processing for LBP images.In order to remove redundant information of each block feature,we use the principal component analysis method to reduce the feature dimension,and finally,we connect the dimension-reduced feature of each block together to form the final face feature.Traditional face recognition retrieval methods often fall into the "dimension disaster"when confronted with massive high-dimensional data,which leads to the low retrieval efficiency.Hash algorithm is able to solve high dimensional large scale data retrieval,In this thesis,we mainly introduce the hash algorithm,including the local sensitive hash algorithm based on random projection,the iterative quantization sensitive hash algorithm based on learning and the enhanced iterative quantization sensitive hash algorithm.In this thesis,we analyze the direction of the retrieval efficiency of these hash algorithms,that is,to reduce the number of data points in the hash buckets,so as to reduce the time of removing repetition,calculating distance and sorting.According to the analysis,this thesis proposes a hash algorithm based pre-classification,which divide the data points into several classes before the hash mapping.First,the linear discriminant analysis method is used to project the high dimensional face feature data to the best classification space,and the feature dimension is reduced.Then,using the fuzzy C means clustering method to deal with the dimension-reduced feature data,the membership degree matrix of the feature data is obtained.The fitness function is established according to the minimum distance between classes and the minimum distance between classes,and the minimum principle of classification intersection.We apply genetic algorithm to optimize the membership threshold.Finally,we classify the feature data intangibly according to the threshold of membership degree.In this thesis,the intelligent face recognition system is developed based on the pre-classified hash algorithm,and we have completed the overall system design,function design,module design and database design.On the windows platform,we use Matlab,C#programming language and Server SQL database to develop the system.We realize the user rights management,the identification of clearance passengers,fugitive identification,record query and other functions.