Study On Depth Image Based Rendering Algorithms

Current graphics research is advancing with the motive of being able to produce photo-realistic images in real-time. Although the growing speed of today's graphics hardware has surpassed the Moore's Law, which has resulted in considerable speed increase in rendering polygonal scenes, the trade-off between rendering quality and speed still bothers graphics researchers. Compared to the traditional geometry-based rendering, Image-based Rendering (IBR) techniques are gaining more attention in graphics community nowadays than ever because of their ability in rendering high quality images with fast speed.In this thesis, we limit our work and discussion in IBR techniques that are based on depth images. Several depth image based rendering methods have been proposed over the last several years. However, we have seen in these methods evident drawbacks like poor image quality, slow rendering speed, huge storage penalty, inadequate representation of the scenes due to deficient image-based modeling schemes, etc. We present a new IBR method, Bar Texture Mapping (BTM), to effectively address these limitations, while at the same time we expect the following advantages in BTM: Support of representation of 3-D surface details and view motion parallax; Easy and efficient implementation on modern graphics hardware; Elimination of hole-filling computation in rendering; Layered Depth Bars (LDB), an extension of BTM after incorporating a LDI's structure, handles complex objects with multi-layered features very well.Chapter 1 first gives an introduction of IBR and traditional geometry-based rendering. In the second Chapter of the thesis, we give a short survey of recent IBR techniques. Several depth image based rendering methods: 3-D Image Warping, Relief Texture Mapping, Hardware-assisted Relief Texture Mapping, Layered Texture Mapping, Layered Depth Images, and Layered Relief Textures are then described. In Chapter 3, we present in detail Bar Texture Mapping, including the preprocessing phase when the bar textures are generated, and two different rendering strategies:Bar-Textures-Assisted Relief Texture Mapping and Occlusion-Compatible-Order Bar Texture Mapping. At last, Layered Depth Bars is introduced to handle complex scenes.To compare the performances, we set up a uniform platform on which all these related IBR techniques are built. Experimental results show higher efficiency in our new approach, especially when multiple factors like image quality, frame rate, and storage overhead are all taken into consideration.