Segmentation of spectral objects from multi-spectral images using canonical analysis

A series of problems in remote sensing require the segmentation of specific spectral objects such as water bodies, saline soils or agricultural fields. Further analysis of these objects, from multi-spectral images, may include the calculation of optical reflectance variables such as chlorophyll concentration, albedo or vegetation humidity. To derive reliable measurements of these variables a precise segmentation - from the rest of image - of the spectral objects is needed. In this work we propose a new methodology to segment spectral objects based on canonical analysis and a split-and-merge clustering algorithm. Three examples are provided to demonstrate the goodness of the methodology.