The Research And Application Of Tensor Locality Preserving Projection Method

In the contemporary, higher-order tensor model gets more and more people’sattention and is being researched in depth. In view of the limitations and the problemof the traditional feature extraction method, this paper puts forward a kind of methodbased on Tensor including supervised and unsupervised local projection algorithm,namely supervised Tensor Local Projection algorithm (STLPP) and unsupervisedTensor Local Projection algorithm (USTLPP). STLPP algorithm only uses the categoryinformation, ignoring the local geometric properties of the data. USTLPP considerskeeping local geometric characteristics of the data, but it ignores the categoryinformation. Considering the shortcomings of the above two methods，we put forwardan improved Tensor Local Projection algorithm (ITLPP)which considers both thecategory information and local geometric characteristics of the data. Through theexperiments on the2D-face databases and3D-face databases based Gabor filter, weverified the feasibility and effectiveness of the three tensor analysis methods.Experimental results also show that the methods based on tensor has higher accuracyrate in the classification recognition than traditional statistical pattern recognitionmethod, and that is superior to traditional methods in terms of computationalcomplexity and storage cost. During the three kinds of tensor method, the recognitionrate of ITLPP algorithm is the highest.Under no prior knowledge, we focus on brain recognition fMRI datapreprocessing, and use improved Tensor Local Projection algorithm for featureextraction, and classified the brain cognitive state. Experimental results show that thepresented method can obtain more accurate recognition rate than PCA in theclassification results...