The Multiprocessor-System-On-Chip (MPSoC) architectures based on the Network-On-Chip (NoC) communication are the one of the most appropriate solution for image and signal processing applications under real time constraints. Due to the ever increasing complexity of these algorithms, the types and sizes of the data manipulated, the MPSoC architectures are necessary to meet the constraints of performance and portability. However exploring the design space of such architecture is time consuming. Indeed, many parameters should be defined such as the type and the number of processing cores, the memory architecture and the communication network between all these components. Validation by high-level simulations has the lack of the precision. Low-level simulation is inadequate for such big size of the architecture. Therefore, the emulation on FPGA becomes inevitable. In image processing, spectral imaging is more and more used. This technology captures light from more frequencies than the human eye increasing the number of wavelengths. Invisible details can be extracted from a scene. The difference between all spectral imaging applications is the number of wavelengths and the precision. Many parameters affect the characteristics of the algorithm, having a huge impact on the final architecture. The objective of this thesis is to propose a method for sizing one of the most accurate hardware and software architecture for multispectral imaging application. The first step is the design of the NoC based on the network traffic. The automatic development of an emulation platform on a single FPGA or multi-FPGAs simplifies this step and determines the positioning of the computational components. Then, the design of computational components and their functions are validated using existing simulation platforms. The synthesizable model of the architecture on FPGA is then generated. The design flow is open. Several NoC structures can be inserted using the source model of this component. The set of results obtained points out the major parameters influencing the performances of architecture and the NoC itself. Several solutions are described and analyzed. These studies allow us to lay the groundwork for a complete MPSoC emulation platform based on NoC
