Fourier Descriptors Based Approach for Hyperspectral Feature Extraction

Taking advantage of Fourier Descriptors attractive properties like geometric transformations invariance and robustness to noise, this work investigates the potential of using Fourier Descriptors as hyperspectral Feature Extraction approach. The main idea is to consider each pixel in the spectral domain as an open linear shape represented by its Fourier Descriptors in the frequency domain. A subset of low order descriptors is retained to represent the overall features of the curve shape, where the very high frequency descriptors that describe the small details of the curve shape are discarded. To evaluate our approach, K-Nearest Neighbor (K-NN) and linear Support Vector Machine (SVM) classifiers are applied to the reduced data resulted from the well-known dimensionality reduction techniques including, Principal Component Analysis (PCA), Independent Component Analysis (ICA) and Minimum Noise Fraction (MNF), and the proposed Fourier Descriptors based approach. Experimental results on real hyperspectral datasets demonstrate that our approach performs the best.