Wavelet Based Time-Varying Volume Representation and Rendering
This project is about compression of time varying volume data suitable
for efficient hardware volume rendering.
While current state-of-the-art hardware allows very fast volume rendering
process, data transfer between disk and main memory or between main memory
and texture memory can become a bottle neck in time-varying hardware
volume rendering due to limited bandwidth of
memory systems. To reduce the size of data set, compression is natural
to apply using temporal and spatial coherence. However, run-time decompression
can result in poor performance which is not suitable for interactive volume
rendering applications.
The contribution of our compression scheme is :
reducing the size of time-varying volume data set with minimal distortion,
Each frame of volume is classified as either base frame or predictive frame.
Assuming that there are only small changes between consecutive frames,
we store only changes from base frame in the predictive frames.
Wavelet transformation is performed on each frames and
coefficients are sorted in decreasing order.
At run-time, we can load as much
coefficients as possible given the time constraints followed by reconstruction. This allows to minimize data read and transfer while keeping best achievable
image accuracy.
Figure 1 : one slice image of base frame volume
Figure 2 : one slice image of predictive frame volume
Figure 3 : one slice image of changes between base frame and predictive frame. compression and decompression of changes are more efficient than those of whole predictive frame.
