| dc.description.abstract | In this thesis I propose a collection of machine vision techniques that could be exploited for the purpose of optically recognizing an integrated circuit (IC) in a semi-interactive manner. It will cover the entire machine vision chain, from the segmentation of the input, to the classification of the extracted information. First the seeded region growing (SRG) algorithm is discussed for segmentation purposes. Following that, I describe how a nearest neighbor classifier and a combination of an out of the box optical character recognition (OCR) solution and the Levenshtein distance can be used to provide input for a new classifier which I named the Intentionally Biased Weighed Voting Classifier (IBWVC). Experiments related to each of the individual topics have been conducted on a small self-assembled dataset. The results show that plain nearest neighbor classification is already rather accurate. Still, improved accuracy can indeed be achieved in a, by assumption, convenient semi-interactive manner by applying the IBWVC. | |
