Exploiting the Context for High Volume Compression of Virtual Reality Data

Virtual reality involves manipulation of massive quantities of three dimensional elevation and imagery data to render a realistic scene as the operator "flies" through. Visualization of the large areas needed in these simulations requires that the data be compressed to very low bit rates to make storage and transfer feasible. An important design requirement for consideration is that the data compression scheme needs to exhibit high fidelity for achieving realism. The primary objective of this Phase I effort is to develop real-time data compression algorithms to support virtual reality data handling. Our approach to achieving high volume data compression that preserves important structures in the data (e.g., discontinuities in an image for draping) makes use of wavelet based algorithms we have developed in a current Phase II SBIR effort supported by NASA. Specifically, we have shown that the generic wavelet compression scheme can be further improved by using information about the type of terrain in an image resulting in compression ratios close to 200:1. Based on our experience with fractals, wavelets and other compression methods, we find that the conjoint spatial- Fourier domain techniques such as wavelets are very effective both from a real-time computational point of view as well as an optimal data compression point of view. hi the case of "draping" a scene for virtual reality, the reduced resolution image set inherent in wavelet based compression will also aid in rendering a perspective image without accessing the entire compressed image.