|
2D transforms have many applications in digital image
processing. In this thesis a programmable 2D transform
processor is presented. By properly load into the transform
coefficients, one can compute all kinds of 2D orthogonal
transoforms. Via table look-up algorithm which is based on
distributed arithmetic, no multipliers are needed. Since the
butterfly stages using in most fast algorithms depends on the
type of transform and result in an irregular structure and the
complicated routing requires large silicon area, we abandon
this structure in our design. The partial sum technique reduce
the number of needed memories, make it possible to realize the
processor in a chip. Under 1.2 micro meter technology, the
simulation shows that this processor can work at the speed of
40 MHz.
|