Footnote:
Diese Datenquelle enthält auch Bestandsnachweise, die nicht zu einem Volltext führen.
Description:
Algorithmic robustness in real-world scenarios and real-time processing capabilities are the two essential and at the same time contradictory requirements modern image-processing systems have to fulfill to go significantly beyond state-of-the-art systems. Without suitable image processing and analysis systems at hand, which comply with the before mentioned contradictory requirements, solutions and devices for the application scenarios of the next generation will not become reality. This issue would eventually lead to a serious restraint of innovation for various branches of industry. This thesis presents a coherent approach to the above mentioned problem. The thesis at first describes a massively parallel architecture template and secondly a seamless simulation- and semiconductor-technology-independent design framework for a class of probabilistic image models, which are formulated on a regular Markovian processing grid. The architecture template is composed of different building blocks, which are rigorously derived from Markov Random Field theory with respect to the constraints of \it massively parallel processing \rm and \it technology independence\rm. This systematic derivation procedure leads to many benefits: it decouples the architecture characteristics from constraints of one specific semiconductor technology; it guarantees that the derived massively parallel architecture is in conformity with theory; and it finally guarantees that the derived architecture will be suitable for VLSI implementations. The simulation-framework addresses the unique hardware-relevant simulation needs of MRF based processing architectures. Furthermore the framework ensures a qualified representation for simulation of the image models and their massively parallel architectures by means of their specific simulation modules. This allows for systematic studies with respect to the combination of numerical, architectural, timing and massively parallel processing constraints to disclose novel insights into MRF models and their hardware ...