University of Houston

Multi-spectral images consist of a series of images each aquired under a narrow band wavelength of light. These images allow to explore spectral differences of objects such as cell smears, and therefore conclude on chemical properties of the objects analyzed. Recently, spectral iamges have been used in supporting medical doctors in early detection and screening of cancer, by using the chemical properties to distinguish between cancer cells and healthy cells. However, the resolution of the images produced by the microscopew combined with the number of spectral channels acquired can lead to overall images ranging from 8 GB - 50 GB making it difficult to analyse these images on a desktop machine.

The objective of this thesis is the evaluation of parallelization strategies for a graph based segmentation algorithm. The resolution of the muli-spectral image and the amount of data that needs to be processed along with the complexity of current image processing algorithms makes it necessary to parallelize the algorithm in order to speed up the analysis of the cell smears. Two approaches of parallelization were implemented using the Message Passing Interface (MPI) standerds. The thesis also presents a comparison between the two approaches based on the quality and performance.