LDA enhanced moments

Moments and functions of moments are powerful tools in a vast number of fields, particularly image signal processing. In this paper, a method for obtaining a set of orthogonal, noise-robust, and distribution-adaptive moments, called Fishermoments (FM), is presented. FM are obtained by performing linear discriminant analysis (LDA) in the moment space resided by geometric moments (GM). The moment space is transformed into the feature space where a separability criterion is maximized. Experiments performed to gauge the performance of FM show significant improvements in terms of accuracy and noise robustness as predicted by the theoretical framework.