Study On Image Warping Theory And Method Of Image Based Rendering

An eternal objective for VR is to achieve real-time, interactive and immersive interoperability. IBR, with the virtue of scene complexity independency and high reality, has become a focusing research area in VR in the recent decade. One of the most practical approaches that used for building IBR navigation system, called depth image based rendering, is widely used for its memory efficiency and high degree of free view point. It generates image for changed viewpoints based on several reference images from nearby viewpoints and their depth information. Image Warping is the core technique for the depth image based new view generation. Study on the theory and practice of Image Warping forms our fundamental incentive for improve VR system performance.This thesis restricts itself on the topic of Image Warping theory and method. Some novel models and algorithms have been proposed to effectively and efficiently render new views based on the existed achievement of previous researchers, which are detailed as follows:(1) It proposes a Zoom Image Warping Algorithm to solve the outside mapping and hole problem in forward image warping. It uses a view shrinking method to decide the scope of valid reference pixels in the corresponding boundary area in the new and reference views. Then with the valid reference region information, it implements template mapping technique to fill in most of the holes. The new algorithm reduces large amount of redundant computation simply by optimizing the rendering order of the object images and mapping direction. Moreover, it avoids the huge requirement of inverse mapping by taking advantage of the template technique, which has improved the computation efficiency without artificial rendering error.(2) It proposes a none-Z-Buffer multi-reference fields warping method. The method eliminates the limitation of using only single reference field in the traditional none-Z-Buffer methods by introducing an "aided view", which is able to decide the global order of all sampled rays from different reference fields. The warping is divided into two-steps to avoid the use of Z-Buffer. The method can be easily implemented based on existing rendering model and pre-processing work. Its visibility correctness has been proved both in theory and in practice.(3) A Crossed Segment Algorithm (CSA) is developed using the restriction relation of the multi-reference fields. In stead of the time consuming process of searching the epipolar line pixel by pixel as in previous approaches, CSA restores new restriction relation among several reference fields with depth information and finds the matching pixels for the new view by solving equations on the epipolar line. For that only edge-pixels not all pixels in epipolar line are deal with, CSA is proved to be of use to accelerate the inverse image warping by experiments.