Hibridni model za segmentaciju snimaka generisanih primenom kompjuterizovane tomografije

Abstract

Kompjuterizovana tomografija (CT) je u poslednje vreme ušla na velika vrata sa razvojem industrijskih CT sistema, usled njene primene u različitim oblastima, a uveliko ulazi i u polje koodinatne metrologije. Zbog karakterizacije objekata sačinjenih od različitih materijala (najčešće metala i plastike), javljaju se određeni problem u vidu nastanka artefakata kod rezultata dimenzionalnih merenja. Istraživanja koja su sprovedena u ovoj doktorskoj disertaciji se bave problemom redukcije uticaja tih artefakata i segmentacije 2D snimaka. Razvijen je novi model koji je baziran na primeni hibridne metode gde je izvršena kombinacija dve metode za obradu slike, a to su fazi klasterizacija i rast regiona. Aksenat je stavljen na primeni ove hibridne metode radi dobijanja tačnijih rezultata segmentacije, što direktno utiče i na rekonstrukciju dimenzionalno tačnijih 3D modela.

Computed tomography (CT) has recently entered a large door with the development of industrial CT systems, due to its application in many different areas, and it is already entering the field of coordinate metrology. Due to its ability to non-destructively characterize objects made of different materials (typicaly metals and plastics), a certain problem arises in the form of artefacts that are present in the results. Research carried out in this dissertation deals with the problem of reducing the impact of these artefacts and the 2D image segmentation. A new model was developed based on the application of the hybrid method where a combination of two methods for image processing was performed, which are fuzzy clustering and region growing. The accent is emphasized in the application of this hybrid method in order to obtain more accurate segmentation results, which directly affects the reconstruction of dimensionally more accurate 3D models.