Investigating Segmentation Robustness on Multispectral Image Color Representation

One major obstacle faced when dealing with multispectral and even more with hyperspectral images is the processing complexity strictly and proportionally linked to the spectral dimension. Projecting the multispectral data on a familiar color space can reduce drastically the resources and the time consumption when investigating the spectral information for segmentation and classification ends. In this paper, we will study the impact of representing a multispectral data in the RGB color space on segmentation outputs. The dataset collected from CAVE dataset will be subjected to dimensionality reduction, then translated to the RGB representation. Watershed segmentation followed by region merging algorithm will be applied to the emerging color portrayal and the results will be assessed regarding precision and visual quality.