A parallelization methodology for reconfigurable systems applied to edge detection

Abstract

In this paper, a novel parallelization methodology is applied to Edge Detection Algorithm (EDA). The proposed methodology is based on a multiprojection approach and on a fusion of processor elements. It eliminates the relationship between problem size and processor array size when using methodologies based on projections. EDA is an interesting problem because its data dependencies and its potential parallelism, besides EDA is used in multiple applications. In this study, multiple versions of the EDA architecture are generated in order to fulfill requirements of throughput and implementation area.