Research On Face Recognition Algorithm Based On Improved LLE And MMC Collaboration

Face recognition is a popular research direction at present,and researchers have proposed many corresponding algorithms for this.Among them,the manifold algorithm can extract the manifold information in the sample.Since it was proposed,it has received widespread attention and has gradually become a hot research direction.Locally Linear Embedding(LLE)as a classic algorithm for manifold learning.It Can maintain the original neighborhood structure in the process of reducing data dimension.It is widely used in various fields such as image classification and pattern recognition.However,the LLE algorithm is sensitive to the selection of the number of neighbor points,cannot effectively use the sample category information,and has a small sample problem.Based on this,this article conducts in-depth research on it.The main research work has the following parts:(1)The LLE algorithm is sensitive to the selection of the number of neighbors,and the Euclidean distance used cannot effectively mine the local structure information.Therefore,an improved Rank-order distance LLE algorithm(Improved Rank-order local linear embedding,IROLE)is proposed.The algorithm first calculates the improved Euclidean distance between each sample point and other sample points,and sorts them by distance.The improved Euclidean distance can equalize the sample distribution,thereby reducing the impact of uneven distribution of sample points on finding nearest neighbors;then calculate the Rank-order distance of the samples based on the sorted distance matrix,and then make full use of the sharing between sample points.Finally,based on the improved Rank-order distance,we can effectively find the nearest neighbors of the sample and construct an objective function with neighborhood information.The experimental results on the ORL and Yale face databases show that the improved local linear embedding algorithm can effectively obtain the neighborhood information of the sample and thus obtain better recognition results.(2)The LLE algorithm cannot effectively use the category information and global information of the samples,and has a small sample problem.Aiming at this problem,a local linear embedding algorithm based on intra-class subspace learnning(LLE/ISL)is proposed.The algorithm first learns the intra-class information to form a new inter-class matrix to reduce the impact of inter-class differences on face recognition.Then it uses the LLE algorithm to extract the manifold information of the samples to obtain the manifold projection matrix.The combination of projection matrices enables the algorithm not only to obtain the intra-class information of the samples,but also to make up for the shortcomings of the LLE algorithm that cannot effectively obtain the global information.Experiments show that the LLE algorithm and the ISL(intra-class subspace learning)algorithm can be complementary,which significantly improves the recognition effect of the algorithm.(3)In order to solve the difficulty of obtaining global sample information and intra-class information in the LLE algorithm,it is combined with the improved MMC(Maximum Margin Criterion)algorithm to propose an improved rank-order local algorithm based on improved LLE and MMC collaboration.linear embedding based on intra-class subspace learning(IROLLE/ISL).This algorithm can improve the sensitivity of the LLE algorithm to parameters and insufficient information about the neighborhood.At the same time,the linear information and intra-class information of the sample can be obtained by combining the MMC algorithm.Realize the algorithm in complex environments,different k values,and different dimensionsFigure[23]Table[4]Reference[58]...