Tools and techniques to aid in the marking and categorisation of digital images

Abstract

Image processing technology is used in everyday applications to do things such as correct red-eye in digital cameras, and to detect terrorists at airports. Recently it has become more widespread and ubiquitous in society. With the surge in use of image processing algorithms and technology, the need for reliable evaluation for those technologies has increased, and research on objective and quantitative performance estimation methods is actively investigated. This thesis starts out by giving a comprehensive overview of image processing algorithms and performance evaluation techniques, ultimately drawing the focus to the area of testing algorithms using ground truth measurements which allow accurate reports on algorithm success and accuracy to be carried out. After understanding this area, it was revealed that no comprehensive tools exist in the industry for the gathering of ground truth data. Thus, the research moved towards finding out what kind of working methodology would best tackle the important area of ground truth gathering through image marking. After this thorough study, a promising framework was proposed, a framework that would certainly fulfil all the requirements of the difficult ground truth gathering task. The thesis then details how the software for ground truth gathering was designed and implemented based on the established framework. The research concludes that the tool developed certainly helped aid the task of algorithm testing, as proven by the testing and use of the application at the partner company.