Iztok Žužek (2012) Visual recognition of coins by computer vision. EngD thesis.
The diploma paper focuses on the issue of computer visual recognition of coins in automatic machines. The recognition falls within the scope of computer vision. The introduction summarizes the history of development of automatic machines and their mechanisms of coins recognition. We came to the conclusion that due to a number of similar coins from different countries, the existing mechanisms should be upgraded by computer visual recognition. We found out that this issue had already been addressed by Fukuni and others in the beginning of nineties of the 20th century. The central part of the thesis summarizes the overview of coin detection methods that were recently published by Chalechale and others. Among these we selected the most appropriate method which met our criteria. We chose the Chalechale's method and theoretically described its functioning, which was subsequently used for the implementation of this method. The method was also tested on two groups of pictures. We tested its effectiveness with respect to correct detection. The results obtained were fairly similar to those of the Chalechale's and Shen's. Later on, we tested the method against the rotation of input images and the number of categories within a data set. We established that, firstly, the rotation of images has an insignificant effect, which proves that the method is invariable to rotation, and secondly that the increase of categories does not cause significant decrease in correct detection effectiveness. The last test verified whether the results could be improved by leaving out the outer coin ring, which usually contains text, however we found out that by doing so the results were insignificantly improved. Despite good results, the method needs many further improvements, some of which we suggested in the conclusion.
Actions (login required)