Face recognition using manifold learning and contourlet transform

Principal component analysis (PCA) is a simple, loss less, fast and efficient method for dimensionality reduction. However, dozens of nonlinear methods such as IsoMap and Locally-linear Embedding have been proposed to tackle the problems related to rampant complex nonlinear data in particular field of machine learning. In this work, we scrutinize and compare PCA and different nonlinear methods for face recognition. in the first step of this method we use Contourlet transform for extracting transformed coefficients from elements of dataset and then seize the advantage of dimensionality reduction by using non-linear methods in addition to PCA method · drawn Outcomes from carried out experiments on real-world face dataset and a novel artificial one delineate that linear and nonlinear algorithms shows almost identical performance and differences in classification rate are trivial to infer which algorithm is dominating. A nonlinearity measure is used to determine the amount of non-linearity of a data collection in the dimensionality reduced subspace. This criterion helps to deduce the effectiveness of nonlinear methods in a logical manner.