JPEG2000–based Compressed Multiresolution Model for Real-time Large- scale Terrain Visualization

This paper presents an effective compressed multiresolution model for interactive large-scale terrain rendering. In contrast to previous approaches, our model enables construction of adaptive terrain triangulation that takes into account local surface features and, at the same time, supports compact representation of the elevation data. The model enables fast extraction of the certain terrain area height field approximation with the specified resolution from the compressed form that is essential for real-time viewdependent rendering. The core structure of the proposed approach is the quadtree. Each quadtree node represents the square patch that approximates some terrain region with the specific accuracy. Each patch is assigned a compact data that is sufficient to incrementally refine its triangulation and height field and obtain the next-finer approximation. Progressive representation of the elevation data uses the JPEG2000 image coder to compress the quantized differences between the values predicted from the coarser approximation and the exact elevations. The algorithm enables constructing compressed representation with the controlled maximum or root-mean-square world space error. This imposes the strict bound on the screen-space error when rendering the simplified model. Due to the high granularity of the LOD selection, the algorithm is not CPU-consuming and fully exploits the power of modern graphics hardware. The algorithm was tested on different data sets; the experiments proved its effectiveness.