Using local preservation projections (LPP), face images are mapped into a subspace of the face for analysis. Different from the main component analysis (PCA) and the linear discriminant analysis (LDA), which effectively only see the Euclidean structure of the facial space, the LPP finds an inlay that preserves the local information and obtains a facial subspace that better detects the essential structure Of the face manifold. Laplacianfaces are the optimal linear approximations to the Laplace Beltrami operator's own functions in the face manifold.
In this way, unwanted variations that result from changes in lighting, facial expression and pose can be eliminated or reduced. The theoretical analysis shows that PCA, LDA and LPP can be obtained from different models of graphs. We compared the proposed Laplacianface approach with the Eigenface and Fisherface methods in three different sets of face data. Experimental results suggest that the proposed Laplacianface approach provides better representation and achieves lower rates of facial recognition error.
Human recognition is essential in a wide variety of applications such as access control, security, surveillance, to name a few. The face is an important feature in identifying a person. The human recognition system operates in a purely abstract way. This presents challenges to computer vision developers to make a machine smart enough to recognize a person's identity through artificial intelligence techniques. Defining the quantitative characteristics of faces from the qualitative and abstract features used by humans poses a great challenge for developers. Facial shapes and related geometry will come to the aid, but precision can be improved through statistical methods.
In object recognition, the abstract features are quantified for optimistic results. The quantification would be complex if the feature set is too large. For unwanted features can be eliminated through PCA. The accuracy and robustness of the results can be improved through Laplacianfaces.