A method and system for removing rain from an image

Abstract

The invention discloses a method and system for removing rain from an image. The method includes: acquiring an image to be derained; inputting the image to be derained into a deraining model to obtain an image after deraining; wherein, the deraining model is obtained by training a deep learning network using training data; deep learning The network includes a first convolutional layer, a long-short-term memory module, a first residual block, a second residual block, and a second convolutional layer connected in sequence; both the first residual block and the second residual block include sequentially connected The third convolutional layer, the fourth convolutional layer and the linear unit function with leakage correction. The invention can improve the rain removal effect of images.

Description

Technical field [0001] The present invention relates to the field of image processing, in particular to an image rain removal method and system. Background [0002] Rainy days, as a common dynamic weather, often interfere with the imaging system. Due to the dynamic and random nature of raindrops, the accumulated rain marks will degrade the image quality, which will seriously affect the performance of outdoor vision systems, such as automatic driving, pedestrian detection and object recognition, etc., so it is particularly important for images to rain. [0003] The current single image de-raining method is mainly divided into two types: the first is based on a priori methods, such as the bilateral filtering method of separating rain marks by decomposing rain images into low-frequency and high-frequency parts; the de-rain method of de-rain and low-rank representation model is established for rain marks to show similar and repetitive patterns in the imaging scene; the rain layer an...

AI Technical Summary

Problems solved by technology

[0004] At present, the image deraining method based on deep learning uses a network with better deraining effect: a double recursive network (DUAL RECURSIVE NETWORK FOR FAST IMAGE DERAINING, DRN network) for fast image deraining, but the network exists as follows Disadvantages: (1) Although the feature map information can be extracted as much as possible through double recursion, the shallower network structure still has the problem of losing feature map information and extracting less feature information, and the rain removal effect needs to be improved; (2 ) The convolutional neural network algorithm adopted by the network, the activation function used in the hidden layer is the Rectified Linear Unit (ReLU) function, the ReLU function will cause the neurons in the neural network to die due to its own reasons, making the neural network The use efficiency of the nodes is low, which will reduce the available information and affect the deraining effect

Images