Parameter Optimization of Approximate Image Processing Algorithms in FPGAs

摘要: Approximate computing has been proposed as a paradigm for contexts where resilience of applications to errors can be exploited, e.g. allowing trade quality off power/energy or hardware resources. Numerous approximation methodologies have introduced in the literature and combining several them result improved benefits. However, techniques require parametrized control loss accuracy, using multiple ones implies explore larger parameter sets. Furthermore, combined methods interact influence error propagation, adding optimization complexity. In this work, we propose an model, targeted multi-objective genetic algorithm, perform design space exploration simultaneously on all parameters each used system. We tailor encoding operations image color processing application so that algorithm converge properly Pareto front with good diversity. The is carried out trade-offs between quality, FPGA resource, power. results show model achieve offers wide range choose from, while taking into account potential interactions techniques.