Title - The Very Fast Curvelet Transform

The Curvelet Transform is an overcomplete digital transformation that
has an m-term approximation error in L2 that is close to optimal for C2
curves.  The current implementation of the algorithm is done in
non-optimized C code that runs fairly slow (the transform takes around 4
seconds for a 512x512 image).  The algorithm will be optimized using
loop unrolling, subword parallel instructions (SSE1), and switching from
double precision floating point operations to single precision floating
point operations.  Speed and reconstruction error results (both
theoretical and experimental) will then be compared. 

-Brian Eriksson
bceriksson@wisc.edu