New shape and texture descriptors for an improved germplasm characterization of the most representative Mediterranean vascular flora, by image analysis technology

The focus of this doctoral thesis concerned the application of Image Analysis technique for an adequate definition of the seed morphocolorimetric parameters, that represents an important diagnostic factor in the plant taxonomy studies and consequently may be of great help for the improvement of the management and the effective ex situ conservation in the germplasm banks. The first part of this study was related to the computer vision fundamentals and statistical treatment of data. Based on the clear concept that the discriminant ability of a classification system depends not only on the intra-specific representativeness of taxa analyzed, but also on the quality and quantity of the parameters measured and used to discriminate between groups of belonging, a new set of recordable shape and texture variables was introduced in a yet consolidated image analysis system, with the aim to improve the performance of the classifiers. In the second part of this dissertation some case studies are reported as applications of this innovative technology, with the aim to prove its great usefulness for systematic purposes. The results here presented confirmed that an extensive database of morpho-colorimetric traits may be applied for taxonomy screening of species groups, comparing with the current systematic, as well as with groupings more recently revealed by genetic studies. The morpho-colorimetric characterization of Cistus L. (Cistaceae) seeds by image analysis was treated: a database of morphometric and colorimetric data was carried out to statistically discriminate and identify both at inter and intra-specific and populational level. The satisfactory discrimination performances reached agree with the results reported in the previous papers on the same taxa, though, the improvement of the image analysis system adopted, in which an overall of 138 seed features was evaluated, allowed to reinforce the discrimination power also when the morphometric variability within each group, such as in inter-population groups, is extremely reduced. The taxonomical differentiation within the Medicago L. sect. Dendrotelis (Fabaceae) was also described. The obtained results confirmed the validity of the proposed method for the taxonomic differentiation of Medicago at specific levels, and its identification capability of regional and populational groups. Furthermore, the relationships among 79 taxa belonging to the Lavatera and Malva genera were tretated, in order to contribute to their doubtful systematic treatment. Finally, this innovative kind of identification system, which method was specifically developed to identify wild seeds, and that requires only a few seconds for scanning and measurement operations, proved to be a quick, repeatable, reliable and non destructive method. It does not require any chemical reagents, expensive analytical consumables or high priced physical preparation of samples, hence it is a very cheap method. This precise and accurate identification system it was only possible thanks to efficient and useful cooperation between taxonomists and image analysis specialists. The expert, practical experience in such different fields allowed the development of a system so complex in its structure and so simple in the use. Indeed, having a broad database of morpho-colorimetric seed features for an adequate amount of families, genera and species would enable the identification of taxa already present in the database. In this way, this innovative tool would open new perspectives in plant taxonomy, but also offer the opportunity for germplasm banks to make identifications in a standard, speedy way. In addition, the availability of morpho-colorimetric data should be helpful for ecological and/or archeobotanical studies such as the prediction of seed persistence in the soil.