An architecture based on reconfigurability and asynchronism for real-time image processing

Abstract

This paper presents the evolution of the Associative Mesh, a massively parallel SIMD architecture based on reconfigurability and asynchronism. To favor its System on Chip implementation, we introduce a reorganization of the structure based on processors virtualization and evaluate its consequences on hardware cost and algorithmic performances. Using an evaluation environment based on a programming library and a parameterized description of the architecture, we show that a virtualized Associative Mesh achieves real-time execution for a number of complex image processing algorithms, including split and merge segmentation, watershed segmentation and motion detection.