Using JPEG2000 compressed images for foveated texture mapping in limited memory environments

Despite advances in hardware technology and software rendering algorithms, there remain several limitations that force design trade-offs in virtual environments and visualization systems, such as the decision between increased level of detail, scene rendering time and the storage size of an object or virtual world. If an application involves the Internet or another restricted bandwidth transmission environment, the amount of data necessary to communicate a high level of detail becomes an even more critical parameter.; Texture mapping and light mapping are used to increase the realism of a scene without increasing the geometric complexity. The cost for these methods is the greatly increased storage for the high quality images used. Texture compression is used to mitigate this drawback, but current techniques are lossy and provide only moderate compression.; This paper proposes a unique solution using images compressed with an algorithm based on the upcoming JPEG2000 standard that reduces the internal memory storage requirements for texture maps in interactive three-dimensional worlds. The Foveated JPEG2000 Texture Mapping method (fJ2k) uses JPEG2000 codestreams decompressed to resolutions that can vary based on current rendering performance. This potentially results in a foveated image, with a greater level of detail at the center of the field of view. The decoding and foveation algorithms are implemented in the graphics-imaging pipeline and have the potential for hardware implementation.; This is an adaptive algorithm for interactive visualization that works to maintain a predetermined target frame rate based on user input and rendering times for an untextured scene. The concept behind the algorithm is to trade image quality for interactivity in situations where the environment is too complex to be rendered in full detail at the target frame rate. A constrained optimization is performed that selects a level of detail to decode images with which to render each visible object to produce the “best” scene possible within a specified frame-time. In contrast to other culling techniques, this algorithm strives to guarantee a uniform, bounded frame rate not by culling model geometry, but by culling texture image detail.