Dimensionality Reduction Algorithms Based On Manifold Learning And Its Application In Face Recognition

The rapid development of information technology provides important support for the research of education technology.The research of information technology promotes the development of education technology research.Adhered the principle,we have studied the update information technology-the dimensionality reduction technology based on manifold learning which is closely related cognitive science.Our new developed algorithm is used for face recognition.Currently,dealing with high-dimensional data is a technological difficulty for the research of information technology,extracting the high-dimensional characteristics of data is an important research hotspot.The unstructured information of high-dimensional data makes the extracted sample data inherently beyond the direct perception ability of people.Therefore,researchers have proposed a number of data dimensionality reduction algorithms,and the dimensionality reduction technology based on manifold learning is one of the most concerned data reduction methods.Manifold learning was proposed at the end of the twentieth century and gradually applied to face recognition,data visualization,graphic retrieval and other fields.Its nonlinear nature,geometric intuitiveness and computational feasibility have yielded satisfactory results on the actual dataset.However,in some universal problems such as generalized learning,supervised learning and large-scale manifold learning,manifold learning still has a lot of research space.In this paper,combined with the research status of manifold learning method and the structural characteristics of high-dimensional data,a series of research work has been carried out from the aspects of algorithm improvement and application(image data and face recognition technology interaction data).The main work and research contents are summarized as follows:(1)Combined with the existing data dimensionality reduction method,this paper studies the typical manifold learning algorithms that are widely used from local retention manifold learning and global maintenance manifold learning.The implementation process,time complexity and geometric characteristics of the algorithm are analyzed and summarized.(2)Based on the in-depth study of the locality preserving projection algorithm(LPP),a clear linear mapping relationship is introduced to construct an attracting vector matrix,and a reformed locality preserving projection algorithm(RLPP)is proposed,which can effectively extract low-dimensional face image information from high-dimensional data and improve recognition rate and recognition speed for face image.The algorithm effectively solves the main problems of the original LPP algorithm:the unsupervised learning method of LPP affects the extraction efficiency of information features when the face pose changes,and can not reasonably use the category information of known data for experimental research.When the measurement sample is subjected to changes in orientation,facial rotation scaling,and jewelry wear,the proposed algorithm is robust to this changes.Experiments show that the improved algorithm's recognition rate is better than LPP,LPANMM and RAF-GE algorithms,which can achieve better than existing improved flow.The shape algorithm has higher precision,strong generalization ability and higher recognition rate,and is not affected by the sample size.