Inpainting makes every sample count

Abstract

In this work we introduce an inpainting design methodology based on masked data acquisition which allows a beneficial shifting of computational load between the data acquisition system and data processing to enhance existing approaches or to find solutions to resource constrained systems. The proposed methodology is a powerful tool to systems, which are limited to, e.g., area or/and energy consumption, but have to handle high data volumes or require long sensing times. Standard methods like JPEG or compressed sensing provide solutions accompanied by a high computational load corresponding to an area and energy consuming large ciruit complexity or insufficient recovered data quality. However, the results in this work based on the proposed inpainting-related guidelines, e.g., for biomedical signal processing or satellite applications outperform State-of-the-Art techniques and prior works without additional computational load.