Control Theory-Inspired Model Design for Single Image De-Raining

Abstract

While existing image de-raining methods have obtained satisfactory performance, lacking theoretical basis prevents them from further improvement. In this brief, we propose an efficient image de-raining network empowered by control theory to address this issue. To the best of our knowledge, it is the first attempt to integrate the control theory into image de-raining model design. Different from previous methods that roughly construct complicated neural networks, our method is theoretical and provides a new perspective for model design. Specifically, by mimicking the signal processing flow of state observable and controllable standard form, we expand them to two network modules, named C-IM and O-IM with every component in the module one-to-one corresponding to each operation involved in state equation. Equipped with C-IM and O-IM, our proposed network could efficiently explore and exploit the features of rain streaks in a recursive fashion. Extensive experiments demonstrate that control theory equipped method is capable of obtaining promising performance and speeding up the model training.