Single-image rain removing method based on deep learning and model driving

Abstract

The invention discloses a single-image rain removing method based on deep learning and model driving. The method comprises the following steps: firstly, preprocessing image data to obtain a rain imageand a corresponding clean rain-free image; then establishing a single-image rain convolution dictionary model and a corresponding optimization problem; designing an iterative solution algorithm onlycontaining simple operation by utilizing a near-end gradient method, decomposing an iterative updating process into a plurality of sub-iterative steps and expanding the sub-iterative steps into network modules in a one-to-one correspondence manner, and establishing a single-image rain convolution dictionary network (RCDNet); the preprocessed rain map is transmitted to RCDNet for iterative training, the RCDNet is iteratively updated through a reverse optimization algorithm, so that the output result of the network gradually approaches the preprocessed clean rain-free map, and a training model is obtained; a to-be-tested rain map is prepared, the trained model is loaded, the rain map is input into the RCDNet for forward calculation, and the output result of the network is the rain-removed image corresponding to the tested rain map. According to the method, the rain removing performance of the current single-image rain removing technology is greatly improved, and obvious interpretabilityand generalization are achieved.

Description

technical field [0001] The invention belongs to the technical field of image processing and deep learning, in particular to a method for removing rain from a single image based on deep learning and model driving. Background technique [0002] Affected by rainy weather, images taken outdoors are usually damaged, such as effective background and texture details will be blocked by highlighted raindrops and dense rain strips, which will seriously affect the performance of outdoor vision tasks, such as object tracking, Video surveillance and pedestrian detection. Therefore, image deraining is a very important preprocessing task that has received extensive attention in recent years. In particular, unlike the video deraining task that can utilize multiple frames of images, single image deraining is a more challenging problem since there is no inter-frame information available. [0003] So far, the existing single-image deraining techniques can be roughly divided into three catego...

AI Technical Summary

Problems solved by technology

For example, the network structure they design is becoming more and more complex, and they do not dig too much into the rationality of its structure and the meaning of the model, but regard it as an encapsulated end-to-end mapping module, so it usually lacks obvious explainability sex

In addition, this type of method basically ignores the inherent prior structure of rain bars, such as sparsity and repetition of local patterns, so it is usually prone to sample overfitting problems, and its generalization performance is limited

Images