An Adaptive Nearest Neighborhood Method And Its Application Of Face Recognition In Access Control

The proposal of natural nearest neighbor(3N)technology effectively overcomes the k-value sensitivity problem in the K-nearest neighbor algorithm and it laid the foundation for the research and development of nearest neighbor technology.As the latest research direction of nearest neighbor technology,self-adaptive neighborhood method has achieved good effect in the fields of clustering,classification and outlier detection.At the same time,the adaptive nearest neighbor algorithm also has its own limitations.As the problem of high time consumption in the iterative search process,the definition of the neighborhood is not accurate enough.This thesis focuses on research and analysis of 3N technology and presents a fast adaptive nearest neighbor algorithm.Then this thesis improved subspace learning algorithm by taking advantage of fast adaptive nearest neighbor algorithm.Finally,the improved subspace learning algorithm is applied to scenes based on face recognition access control system.The main work of this thesis includes the following aspects.Firstly,for the natural neighbor algorithm,there is a problem that the search time of data points has high complexity and the number of neighbors in the nearest neighbor is not uniform.According to the idea of the natural neighbor algorithm,this thesis uses the most remote neighbors of the reverse search data to achieve the saturation of the data point neighbor graph.Experiment shows,a strictly confirming rule with fast searching proposed in this thesis takes less time than the natural neighbor algorithm for classic face database ORL and GT data.Secondly,the MFA algorithm is an excellent algorithm in subspace learning.It explores the statistical characteristics of data distribution through the geometric relations between neighboring data points.For the problem that the number of neighboring data points is difficult to select,this thesis proposes an MFA algorithm based on fast adaptive nearest neighbor algorithm,which automatically selects the number of neighbors to avoid the problem of parameter selection.Experiments show that the MFA algorithm based on adaptive nearest neighbor proposed in this thesis effectively improves the classification accuracy of the MFA algorithm and also makes the MFA algorithm more intelligent.Thirdly,based on the research of face recognition related technologies,this thesis describes the working principle of the system and the realization of the main functional modules and a simple intelligent face recognition-based access control system is designed.