Discrete optimization methods and their applications in computer vision

点击翻译按钮获取中文摘要

Computer vision, simply stated, is about teaching computers to see, that is to perceive which objects are present in a scene, where they are located, and where they are moving. The ability to atomatically process visual information leads to a virtually unlimited number of applications, ranging from the traditional ones such as industrial inspection, security and surveillance, robot navigation, to the newer ones such as video conferencing, virtual travel, and special effects for the movie industry. New applications are being developed each year.An optimisation based approach to computer vision is particularly promising and has already proved quite successful in applicability across different vision domains. There has been a tremendous progress in recent years in such fundamental computer vision tasks as stereo correspondence, image segmentation, image restoration, multi-view reconstruction, in large part due to the development of effective optimisation techniques. It should be noted that the recent progress is due to the optimisation algorithms themselves, since most of the objective functions they are being applied to have been used in computer vision for a long time but without good means to optimise them, and, therefore, with mediocre performance. A large portion of my previous research has been devoted to developing efficient optimisation algorithms, designed specifically for computer vision problems. At the same time, I am very interested in applications of optimisation methods to specific computer vision tasks. I find that there is a synergy between the theoretical methods and practical applications. It is from trying to solve a practical application that one finds interesting new optimisation problems. In turn, solving a new optimisation problem leads to an improved  performance on a practical application. I propose two research efforts, one directed at the theoretical issues in optimisation based techniques, and the other direction is towards practical applications of optimisation techniques.  The research in these two directions is closely intertwined.