the optimum system implementation. On the other hand,
this automatic system generation must not require specifications
which are too coarse grained for efficient system implementation.
Determination of the smallest possible buffer
size for instance requires a model on pixel level [11]. In
order to fulfill these requirements, this paper presents a new
approach, how image processing algorithms can be modeled
in SysteMoC [4], a SystemC-based modeling language.
It allows to generate executable specifications together with
the capability for analysis and optimization. Its key benefits
are the capability to (i) represent and (ii) classify static and
data dependent one- and multi-dimensional algorithms and
(iii) the possibility to distinguish between point, local and
global algorithms thus allowing for efficient buffer analysis
and implementation. This does even apply for nonrectangular
windows.