A random forest-based method for selection of regions of interest in hyperspectral images of ex vivo human skin
Journal article, Peer reviewed
Published version
View/ Open
Date
2019Metadata
Show full item recordCollections
Abstract
Hyperspectral imaging is a useful tool for characterization of human tissue. However, the vast amount of data created makes it challenging and tedious to manually select spatial regions of interest for further processing. In this study, a random forest-based method was evaluated on basis of its ability to segment human skin regions from the background. The method was compared to the performance of two alternative methods, spectral angle mapper (SAM) and a K-means clustering-based method. The methods were tested on hyperspectral images of ex vivo and in vivo human skin in the wavelength range 400-1000 nm. The random forest approach was found to be robust and perform well regardless of image type. The method is simple to train, and requires minimal parameter tuning for good skin segmentation results.