Mahalanobis Distance Metric Based Laplacian Mapping for Image Recognition

摘要

An Improved algorithm for image recognition, called Mahalanobis Distance Metric based Laplacian Mapping Algorithm(MLMA), is presented in this paper. Firstly MLMA learns a Mahalanobis metric matrix from training samples, then we use the Mahalanobis metic as a similarity measure in Laplacian Mapping Algorithm. Comparison of MLMA and standard Laplacian Mapping Algorithm in ORL and USPS databases proves that MLMA is more effective and robust than standard Laplacian Mapping Algorithm.