Method, device, equipment and computer-readable storage medium for removing rain from video

Abstract

The invention provides a video rain removal method, device, equipment and computer-readable storage medium. The method comprises: building a cyclic deep neural network model for video deraining; training the cyclic deep neural network model to obtain an optimized cyclic deep neural network model; adopting the optimized cyclic deep neural network model to treat rain The video performs the rain removal operation. Because the cyclic deep neural network model can encode the background information between adjacent frames, extract more representative features, and integrate redundant information in the temporal and spatial domains of the rainy video, it can achieve a good video deprecation. The effect of rain makes the restored video more accurate and clear.

Description

technical field [0001] The embodiments of the present invention relate to the technical field of video processing, and in particular, to a method, device, equipment, and computer-readable storage medium for removing rain from a video. Background technique [0002] Videos shot under severe weather conditions usually face a series of problems such as image blur and image content coverage. As the most common factor affecting rainy videos, raindrops usually cause occlusion and blurring of local areas. This signal distortion and loss of detail can impact the performance of many outdoor vision applications that use high-quality video as input. Therefore, the detection and removal of raindrops in videos is of great significance. [0003] Existing video deraining methods utilize information redundancy in the spatial and temporal domains, and methods based on physical modeling, such as using the orientation and color characteristics of raindrops for video deraining. Or use dynamic...

AI Technical Summary

Problems solved by technology

[0005] Embodiments of the present invention provide a method, device, device, and computer-readable storage medium for removing rain from a video, which solves the technical problem that the video restored by the method of removing rain from a video in the prior art is not accurate and clear

Images