Static and Dynamic Image Analysis and Synthesis

点击翻译按钮获取中文摘要

Computer vision (CV) focuses on using mathematical methods to extract the shape, structure, semantic information or other intrinsic properties of a scene. The goal of computer graphics (CG) is the reverse process of generating desired images, photographic or artistic, using computer algorithms. CV and CG are intimately related. Hence, research results in one area can often be applied or extended to the other area. My long term research goal is to develop a process or processes that can facilitate the interchange of results between CV and CG. During the last 5 years, when my students and I were developing an underwater 8-camera array system for Neptune Canada, we discovered that many previous works in underwater imaging have simplified incorrectly the refraction effect. Since I am developing the next generation of underwater 3D vision system with Venus Canada and since there is very little work in physically correct underwater computer vision algorithms, I plan to focus my efforts on issues related to underwater imaging by developing physically correct methods. In the short term, I plan to focus on the following objectives within the next five years: 1) to develop a physically correct framework for developing underwater computer vision algorithms, 2) to use the framework to develop physically correct underwater computer vision algorithms, in particular, camera calibration and 3D reconstruction, 3) to use the framework to develop new algorithms in synthetic aperture imaging and in structured light techniques for the underwater environment, and 4) to extract attributes or features that can help to animate fluid more realistically. To achieve objective 1, I plan to investigate different mathematical frameworks that can address the issues of refraction. The goal is to develop a set of mathematical tools similar to what has been done for land-based systems. For objective 2, I plan to use the framework developed in objective 1 to derive select computer vision algorithms, in particular, camera calibration and 3D reconstruction. As well, I will demonstrate the validity of these algorithms experimentally. The next problem that I will investigate is in 3D reconstruction. The success of land-based computer vision algorithms is due to the discovery and the use of constraints, e.g. epipolar, optical flow, left-right consistency, to name a few. In the underwater environment, I will extend constraints used for land-based system to the underwater environment and to develop new ones. The plan for objective 3 is to investigate two major computer vision topics, namely, synthetic aperture imaging (SAI) and structured light. For SAI, I will extend my recent work to develop one for the underwater environment. One advantage of SAI is its ability to see through occlusion, which is very useful when some portion of a camera is occluded. An underwater structured light system will provide much improvement to the accuracy of the correspondence process. Realistic rendering of fluid is closely related to the above mentioned work. On land, using computer vision algorithms, extracting the intrinsic properties of a scene can enhance the quality of rendered images. By the same token, for the underwater environment, the extracted attributes of the underwater environment will enhance the quality of rendered images. Since there is very little work done in underwater computer vision and computer graphics, my work will certainly have significant short as well as long term impact to the research community.