Research On Linear Dimensionality Reduction Algorithm And Its Application In Face Recognition

With the development of computer technology,variety of areas such as computer vision,face recognition,biological information and medicine are facing with high-dimensional data problems.High-dimensional data often contains redundant information and noises.Dimensionality reduction is to project high dimensional data into low dimensional space,which keeps the original data structure simultaneously.In the past few decades,it was studied continuously by many scholars and was made great progress.In face recognition,face images are often difficult to identify due to the influence of light,expression and posture.The dimensionality reduction can extract the effective features in face images,remove the redundancy and interference,and improve the recognition rate.The main research contents of this paper include the Adaptive Average Neighborhood Margin Maximization for Dimensionality Reduction(AANMM)and the application of the algorithm in face recognition.The details are as follows:(1)In order to overcome the problem of Euclidean distance can't precisely measure the similarity of the data which contains useless feature and noise in the pattern recognition research area,a novel adaptive average neighborhood margin maximization(AANMM)for dimensionality reduction method has been proposed.The proposed method learns an optimal subspace iteratively,then selects k nearest neighbors for each target sample in this subspace.Since the useless feature and noise have already been removed so that selecting the nearest neighbors in the optimal subspace is more accurate than in original space,which improves the robustness of the model.Experimental results indicate that our method outperforms conventional methods on diverse real data sets including six UCI data sets and Coi120 object data set.(2)In order to extract face features more effectively,we apply the AANMM algorithm to face recognition.We compare and experiment with several other traditional algorithms on YALE and AT&T face data sets.Experimental results indicate that our algorithm has a higher average recognition rate in face recognition.