Chapter 3: Segmentation

Title: Region growing segmentation of diffused hypercubes in imaging spectroscopy

Author(s): Marcos Ferreiro-Arman, Julio Martin-Herrero

Address: Department of Signal Theory and Communications, ETSIT, University of Vigo, E36310 Vigo, Spain | Department of Signal Theory and Communications, ETSIT, University of Vigo, E36310 Vigo, Spain

Reference: Atlantic Europe Conference on Remote Imaging and Spectroscopy pp. 91 - 95

Abstract/Summary: We present a segmentation method for hypercubes in imaging spectroscopy based in anisotropic diffusion and seeded region growing. Hyperspectral images with a large number of highly correlated bands allow dealing with the spectral axis as a third spatial dimension, such that 3D diffusion methods can be used instead of multivalued or vectorial diffusion methods which involve as many dimensions as spectral samples (there is no spectral axis for them), with the subsequent increased computation time and parametric complexity. We use 3D diffusion with differentiated tuning of the diffusion parameters for the spatial and spectral axes, and we get improved results in smoothness of the filtered images. A reliable smoothing process is critical for robust region growing in hypervalued images. On the filtered hypercube we apply a seeded region growing process based on Euclidean distances between the smoothed spectral signature of the current pixel and the average spectral signature of the growing region. The final application of the segmentation determines the quality of the results.

Order a copy of this article