Point-based modeling for effective rendering
Arie E. Kaufman, Haitao Zhang · 2006
We present methods for point-based modeling for effective rendering. The use of point as a modeling and rendering primitive has become very popular in recent years. Unlike a triangular mesh, a point model does not maintain consistent connectivity information. This leads to certain flexibility in the modeling and processing, and becomes a challenge in the rendering, because the pipeline of the graphics processing unit (GPU) is optimized for the rendering of the triangle primitive. We present several approaches to model objects using point primitive for effective rendering on modern GPUs. Range scanning is an important method for model acquisition. The point model is generated by combining together several range images of an object scanned from different viewpoints. We present a registration method based on probability field to combine points from different range images. Adding textures to a model can enhance the visual realism of the resulting rendering. A hybrid texturing algorithm using both parameterization method and neighborhood matching method is introduced to color the point model by sample textures. Our rendering methods are based on surface splatting, a widely used method for direct rendering of point models. In order to render scenes with a large number of objects, we model every object as a uniformly-sized point hierarchy and use a classification-based rendering. We further introduce the point-and-edge model for edge preserving rendering. A point-and-edge model consists of both surface points and sharp edges. In our constrained splatting algorithm, the projected splats are clipped against the sharp edges. In addition, we introduce a real-time point-based non-photorealistic rendering (NPR) method to generate a line drawing of point models, which supports for level-of-detail control and frame-to-frame coherence. We also use the point primitive for isosurface representation and rendering. Two point-based isosurface exploration methods are presented: edge splatting and an edge kernel method. These two methods integrate point generation and rendering together and are implemented on the GPU for interactive isosurface exploration and high quality rendering.