A Distribution-Dependent Mumford–Shah Model for Unsupervised Hyperspectral Image Segmentation

Hyperspectral images provide a rich representation of the underlying spectrum for each pixel, allowing pixel-wise classification/segmentation into different classes. As acquisition labeled training data is very time-consuming, unsupervised methods become crucial in hyperspectral image analysis. The spectral variability and noise make this task challenging define special requirements such methods. Here, we present novel segmentation framework. It starts with denoising dimensionality reduction step by well-established Minimum Noise Fraction (MNF) transform. Then, Mumford-Shah (MS) functional applied to segment data. We equipped MS robust distribution-dependent indicator function designed handle characteristic challenges To optimize our objective respect parameters which no closed form solution available, propose an efficient fixed point iteration scheme. Numerical experiments on four public benchmark datasets show that method produces competitive results, outperform three state-of-the-art substantially these datasets.