ARCHIVES

The Minimum Noise Fraction (MNF) transform is a widely used method for noise reduction in hyperspectral imaging; conventional one-dimensional noise estimation techniques fail to consider correlation among neighboring pixels, reducing their ability to model noise effectively. Additionally, the standard MNF approach often struggles with low-variance spectral bands due to singular covariance matrices. To address these challenges, this work presents a new Python-based MNF library that introduces a "Three-Pixel" noise estimation method that captures noise in multiple spatial directions and incorporates an adaptive regularization scheme for stable covariance inversion. The experimental results on hyperspectral imagery demonstrate that the proposed strategy consistently improved noise suppression in more that 95% of spectral bands.