Survey of polygonal surface simplification algorithms on GPU

Rendering time depends on number of faces in polygonal mesh. Displaying large number of objects or objects with a lot of faces can result in performance degradation. One of the ways to overcome this problem is to use several levels of detail for polygonal mesh. Levels of detail with low quality are used when object is far away from camera and difference in number of faces is not apparent to viewer. Coarse meshes can be generated manually which is difficult and time consuming. For many applications (computer graphics, computer vision, finite element analysis) it's preferable to use automated methods for polygonal surface simplification to increase performance. Traditional methods for surface simplification are designed for CPU. GPU methods provide improvements in performance and have comparable visual quality. This survey covers algorithms on GPU. They are based on either shader programming or general purpose computing frameworks like OpenCL and CUDA. Shader methods are very restrictive. They are designed to render images in real-time. General purpose computing frameworks provide flexible APIs but require attention to hardware implementation details to avoid performance bottlenecks. Main features of algorithms are provided for comparison as the result. The most used operation for geometry simplification is edge collapse. It's difficult to avoid expensive data exchange between GPU and CPU. It's important to take into account computational model of GPU to increase number of stream processors working in parallel.