Morpho-Colorimetric and Non-Parametric Analyses in Statistical Classification of Vascular Flora (Classification in Image Analysis)

Luca Frigau Abstract of PhD thesis This dissertation deals with statistical methodologies to apply to morphological classification of seeds through extracting information directly from their digital images. It concentrates more on the classifi- cation task, trying to enhance the quality of prediction, and on the automatizing of the classification process. These tasks are very important in botany because they avoid human contradictions in seed classification and to save a lot of time to specialized botanists. Firstly we focused on describing all stages necessary to move from a picture containing raw information of scanned objects to a data matrix usable as input for further statistical analyses. We illustrated how to convert an image so as to enhance its inner contrast in order to get easier the image segmentation. It has been introduced an approach that adapts a widely used method for detecting moving objects from video, called background subtraction (foreground detection), to image segmentation framework. It has been shown how it assists segmentation process to get good results, and allows to automate the process when foreground color of images is not constant, as well as speeding it up significantly. Then methods for enhancing quality of objects and removing residual noise have been illustrated. At the end of..