Robust Feature Extraction Algorithm Based On Noise And Its Application

In recent years,image recognition technology has been widely used in many fields,such as computer vision,pattern recognition,etc.In the process of image recognition,feature extraction will directly affect the accuracy of image recognition.Many existing algorithms can achieve better results only when the image noise is small,but in the actual scene,the image often exists a variety of noise,such as illumination,shadow,occlusion,etc.When faced with these images,the power of the algorithm is greatly reduced.Therefore,based on the characteristics of low-rank regression and low-rank representation,and combined with manifold learning and non-negative matrix factorization,this paper proposes two relatively robust feature extraction model algorithms for occluded images.The specific research is as follows:Locality preserving projection(LPP)has been widely used in feature extraction.However,LPP does not use category information of data,and uses ₂L-norm for distance measurement,which is highly sensitive to outliers.In this paper,we consider the weight matrix of LPP from a supervised perspective,and combine the method of low-rank regression to propose a new model called supervised low-rank embedded regression to discover and extract features.By using L_2,1-norm to constrain the loss function and the regression matrix,not only the sensitivity to outliers is reduced,but also the low-rank condition of the regression matrix is restricted.Then we propose a solution to the optimization problem.Finally,we apply the method to a series of face database,hand-writing digital dataset and palmprint dataset to test performance,and the experimental results show that this method is effective by comparing with some existing methods.Unlike LPP,which maintains local internal structure,Nonnegative Matrix Factorization(NMF)strengthens the local characteristics of data by constraining the non-negative properties of elements.However,because LPP and NMF pay more attention to the local information of the data,it ignores the global representation of the data.In terms of image classification,the global information of the data is often more robust to noise than the local information.So,in order to improve the robustness of the algorithm,we combine with the data of local and global representation,and consider the characteristics of low-rank representation.This paper proposes a nonnegative supervised low-rank embedded model.This method assumes the existence of noise in the data,decomposes the data into clean data and noise data,and makes sparse constraints on the noise matrix through the ₁L-norm,so as to enhance the robustness to noise.In addition,the method uses low-rank representation to learn a low-rank matrix,and then through non-negative decomposition,the robustness of the algorithm is enhanced again.Finally,combined with the graph embedding theory in Study I,the locality of NMF is enhanced while the global data is retained.We also apply the proposed this method to a variety of noise-adding databases and test the robustness of the proposed method.