Parallel Segmentation of Multi-Channel Images Using Multi-Dimensional Mathematical Morphology

Abstract

Multi-channel images are characteristic of certain applications, such as medical imaging or remotely sensed data analysis. Mathematical morphology-based segmentation of multi-channel imagery has not been fully accomplished yet, mainly due to the lack of vector-based strategies to extend classic morphological operations to multidimensional imagery. For instance, the most important morphological approach for image segmentation is the watershed transformation, a hybrid of seeded region growing and edge detection. In this chapter, we describe a vector-preserving framework to extend morphological operations to multi-channel images, and further propose a fully automatic multi-channel watershed segmentation algorithm that naturally combines spatial and spectral/temporal information. Due to the large data volumes often associated with multi-channel imaging, this chapter also develops a parallel implementation strategy to speed up performance. The proposed parallel algorithm is evaluated using magnetic resonance images and remotely sensed hyperspectral scenes collected by the NASA Jet Propulsion Laboratory Airborne Visible Infra-Red Imaging Spectrometer (AVIRIS).