Single Image Deraining: From Model-Based to Data-Driven and Beyond

Understand

The goal of single-image deraining is to restore the rain-free background scenes of an image degraded by rain streaks and rain accumulation.

• The early single-image deraining methods employ a cost function, where various priors are developed to represent the properties of rain and background layers.
• Since 2017, single-image deraining methods step into a deep-learning era, and exploit various types of networks, i.e.
• convolutional neural networks, recurrent neural networks, generative adversarial networks, etc., demonstrating impressive performance.