A Vector Median Filter For Hyperspectral Images Based On Lexicographic Ordering of Estimated Auto-Correlation Functions

We propose a vector median filter for hyperspectral images based on a ranking of pixel spectral values. The ranking of the pixel values is performed in the Fourier domain, based on the hypothesis that these values represent power spectral density functions of unknown random processes, under the assumption that these underlying random processes are stationary. The employed ordering scheme is a classical lexicographic one. We analyze the displacement maps and compare the proposed vector median filter against a classical vector median filter extended to the hyperspectral image case. We present and discuss the experimental results and then draw conclusions.