Actor-Oriented Modeling and Simulation of Sliding Window Image Processing Algorithms

Embedded real-time image processing systems have to process huge amounts of data with limited resources and energy. Hence high efficiency is not only required for manual, but also for automatic system generation. Therefore, in order to allow for different optimizations, a system specification must be such that important algorithm properties are accessible to the system design software. In this paper, we present a new method how multidimensional image processing algorithms can be modeled by actor-oriented dataflow semantics. Using the example of a binary morphological reconstruction, we investigate the modeling requirements posed by point, local and global image processing algorithms. We show how they can be taken into account in our approach, so that efficient implementation and analysis in terms of buffer size and throughput is possible. In particular, by the explicit specification of the communication behavior, both static and data dependent algorithms are supported allowing for a complete system specification.