Mean-shift clustering for interactive multispectral image analysis

Abstract

Mean shift clustering and its recent variants are a viable and popular image segmentation tool. In this paper we investigate mean shift segmentation on multispectral and hyperspectral images and propose three new algorithms. First, we improve segmentation performance by running mean shift on the spectral gradient. At the same time, we adapt a popular superpixel segmentation method to the multispectral domain using modified similarity measures from spectral mapping. Based on superpixels, we design two mean shift variants that both obtain competitive segmentation results in significantly reduced running time. For one variant, the speedup in our benchmark is over 100 times. This enables mean shift clustering in an interactive setting.

A robust kernel density estimator based mean-shift algorithm

“…|x k+1 l − x k l | < . Note that the difference between (14) and (15) are the weights w l . The formulation of the robust mean-shift vector, thus, results in a separation of the problem, where the KDE is, first, expressed robustly, hereby down-weighting the outliers.…”

“…There has been some work on the median-shift as a more robust alternative to the mean-shift that additionally achieves a significant speed-up [12,13]. Nevertheless, it tends to under-segment the data [14], and therefore, does not always correctly assign each data point to the corresponding cluster.…”

Classification of vertebral column disorders and lumbar discs disease using attribute weighting algorithm with mean shift clustering

“…Jordan and Angelopoulou examined performance of the mean shift segmentation algorithm in multispectral and hyperspectral images and suggested three new algorithms. They improved the segmentation performance by making the mean shift algorithm in spectral gradient work [3]. Zhou et al suggested an area-based image segmentation method with the mean shift clustering algorithm.…”

A Novel Framework for Interactive Visualization and Analysis of Hyperspectral Image Data

“…(i) Interactive visualization of spectral distributions based on efficient parallel coordinates [6] (ii) Supervised segmentation of hyperspectral data [10] (iii) Fast global clustering with superpixel support [11] (iv) Fast nonlinear false-color visualization [12] These methods were derived by adapting established algorithms for hyperspectral data and interactive time constraints. They allow us to introduce new paradigms in hyperspectral image analysis that focus on interactive raw data exploration, built on the tight incorporation of the aforementioned techniques into a comprehensive open-source software framework.…”