A multiprocessor architecture for two-dimensional digital filters (computer, signal, processing, image, parallel)
Jung Hyoun Kim · 1985
Multiplication, addition and data transfer are the primary operations required for digital filtering. It is important to minimize data communication requirements (data transfer) as well as exploit inherent parallelism for algorithms in order to implement them in real-time with the use of a multiprocessor system. This research is oriented toward the development of a computationally efficient algorithm for digital filtering. A state-space representation is derived for general order one-dimensional and two-dimensional digital filters. An algorithm is then developed from this representation. This algorithm not only exposes the inherent parallelism for these filters but also reduces data communications requirements. A software package for spatial domain IIR digital filters has been implemented on VAX-11/780. This software package is based upon the state-space representation and was used to filter various images to test the performance of the algorithm. A new generic computational primitive is introduced for the implementation of any arbitrary order one-dimensional or two-dimensional FIR or IIR digital filter. A computational structure based upon this primitive has very high efficiency and very low data transfer and storage requirements. It can form the basis for a programmable special purpose chip for digital signal and image processing applications. A new multiprocessor architecture for real-time image processing is developed with each processing unit in the network implementing the computational primitive. This multiprocessor system has a simple control scheme and a simple interconnection network. Thus, it avoids the bottleneck associated with traditional parallel computers and multiprocessor systems.