Multi stage image restoration based on a single image

A series of recent works based on convolutional neural networks have successfully captured raindrops. However, the specific difficulties of recovery still exist. The task of removing rain streaks from images requires an effective balance between various image details and contextual correlations of image information. This article proposes a multi-stage image restoration network CKL-Net based on a single image to effectively restore the initial background texture in rainstreaked images. Unlike traditional models for removing rain streaks in some encoder decoder images nowadays, our network structure is mainly a multi-level stage architecture, allowing our model to learn the image by stage based on the information contained in the image and gradually restore the background texture of the image, Each stage contains different functions to break down the entire recovery process into more manageable steps, maximizing the utilization of image information to improve performance without additional branches.