Statistical Techniques and Artificial Neural Networks for Image Analysis

The main topic of this PhD thesis is image analysis. The subject was investigated from different perspectives, starting from different image types and with different goals. At the beginning x-ray hazelnuts images were analyzed. The target of this analysis was to determine whether a hazelnut was good or not. In order to do this an Artificial Neural Network was used, whose features were statistical variables. At a later stage a SVM was implemented to try to get better results. The second kind of images were still x-ray ones; they were, however, images coming from a PCB productive process. What we were asked to do was to highlight the air bubbles trapped into a solder joint (particularly those inside the thermal pads). In this case filters and morphological operations were used. The third case were ulcers photographs: the goal of the collaboration with SIF (Societa Italiana di Flebologia, Italian society of phlebology) is to give doctors a way to evaluate ulcers remotely, in order to customize the treatments according to how the healing behaves. A small digression was the development of a small and cheap Arduino-based robot for an educational laboratory (Xke?, in collaboration with prof. Angelo Raffaele Meo from DAUIN, Politecnico di Torino). This should have been the first step towards the development of an evolved robot for agricultural purposes, but the project didn't start