SUPER-FAST
ALGORITHMS OF DISCRETE ORTHOGONAL TRANSFORMS AND PARALLEL-RECURSIVE CONVOLUTION
CALCULATION
New DFT algorithms
combine the use of familiar reduction schemes (Cooley-Turkey, Good-Òhîìàs,
Vinograd) with a special data representation in recurrent number systems
in the algebraic number fields.
Convolution algorithms
rely on:
à) new DFT algorithms and optimal partitioning,
b) parallel-recursive convolution calculation at polynomial approximation
of impulse characteristics (convolution kernel).
 |
 |
|
Superfast
algorithms of spectral transforms
|
Optimal
convolution partitioning
|
Parallel-recursive convolution implementation
 |
 |
|
Polynomial
bases
|
Example:
: selection of convolution algorithms with Gaussian kernel
|
Publications:
1. Proc. of the 7th Intern. Conf. on Comp. Analysis of Images and Pattern,
Springer. LNCS 1296, pp.621-628.
2. Pattern Recognition and Image Analysis, 1998, v.8, N3, pp. 347-349.