Traffic image deblurring method and device based on multi-scale adversarial learning

Abstract

The invention discloses a traffic image deblurring method and device based on multi-scale adversarial learning. The method comprises the following steps of: 1, generating a clear-fuzzy traffic image pair by adopting a motion blur generation method; 2, constructing a traffic image deblurring network model based on adversarial learning, and carrying out adversarial learning on the recovered image to enable the image to be closer to a real clear image; Step 3, constructing a plurality of image deblurring network models with the same structure and the scales from small to large, transmitting a clear image recovered under a small scale as input to a subsequent network, recovering scale by scale, and finally obtaining a deblurring image of the original size; And step 4, carrying out joint training on the multi-scale adversarial learning image deblurring network model. According to the method, the quality of the deblurred image can be improved through a multi-scale adversarial learning method.

Description

technical field [0001] The present invention relates to the field of artificial intelligence technology and image restoration, in particular to a traffic image deblurring method and device based on multi-scale adversarial learning, mainly through adversarial learning on multi-scale deblurred images, so as to improve the blurring performance of vehicles in traffic images Recovery effect. Background technique [0002] With the extensive development of artificial intelligence research and the large-scale popularization of related applications in the past decade, intelligent transportation has become a hot research field in recent years. Traffic images have a wide range of practical applications such as license plate recognition, vehicle counting, etc. However, in the traffic scene, due to the relative motion between the driving vehicle and the fixed monitoring equipment, there is a certain degree of blurring in the image, and the blurred image will affect the subjective vision...

AI Technical Summary

Problems solved by technology

[0005] Aiming at the problem of poor detail information restoration in the current mainstream image deblurring methods, the present invention proposes a traffic image deblurring method and device based on multi-scale confrontation learning

Images