3D Mesh Simplification Techniques for Enhanced Image Based Rendering

Three dimensional videos and virtual reality applications are gaining wide range of popularity in recent years. Virtual reality creates the feeling of 'being there' and provides more realistic experience than conventional 2D media. In order to feel the immersive experience, it is important to satisfy two important criteria namely, visual quality of the video and timely rendering. However, it is quite impractical to satisfy these goals, especially on low capability devices such as mobile phones. Careful analysis of the depth map and further processing may help in achieving these goals considerably. Advanced developments in the graphics hardware tremendously reduced the time required to render the images to be displayed. However, along with this development, the demand for more realism tend to increase the complexity of the model of the virtual environment. Complex models require millions of primitives which subsequently means millions of polygons to represent it. Wise selection of rendering technique offer one of the ways to reduce the rendering speed. Mesh-based rendering is one of the techniques which enhances the speed of rendering as compared to its counterpart pixel based rendering. However, due to the demand for richer experience, the number of polygons required, always seem to exceed the number of polygons the graphics hardware can efficiently render. In practice, it is not feasible to store large number of polygons because of storage limitations in mobile phone hardware. Furthermore, number of polygons increase the rendering speed, which would necessitate more powerful devices. Mesh simplification techniques offer solution to deal with complex models. These methods simplify unimportant and redundant part of the model which helps in reducing the rendering cost without negatively effecting the visual quality of the scene. Mesh simplification has been extensively studied, however, it is not applied to all the areas. For example, depth is one of the areas where general available simplification methods are not very well suitable as most of the methods do not consider depth discontinuities very well. Moreover, some of the state of the art methods are not capable of handling high resolution depth maps. In this thesis, an attempt is made to address the problem of combining the depth maps with mesh simplification. Aim of the thesis is to reduce the computational cost of rendering by taking the homogeneous and planar areas of the depth map into account, while still maintaining suitable visual quality of the rendered image. Different depth decimation techniques are implemented and compared with the available state of the art methods. We demonstrate that the depth decimation technique which fits the plane to depth area and considers the depth discontinuities, outperforms the state of the art methods clearly