A promising power-saving technique: Approximate computing

Abstract

Approximate computing is introduced as an important low-power technology for image processing in recent years. Since a slight decrease in image quality is normally acceptable by human eyes, approximate computing optimizes design by allowing some tolerable errors and sacrificing some accuracy in the computational process. The reduction of computing complexity can thus contribute to the improvement of circuit energy efficiency. This paper reviews existing approximate techniques in image processing field and classifies them into algorithm level, logic level and circuit level. In addition, this paper analyses and highlights the merits of each technique.