基于生成对抗网络的交通模糊图像复原

Restoration of traffic blurred images based on improved GAN

针对光线及车辆运动造成的交通图像模糊和分辨率低等问题,提出一种基于生成对抗网络(GAN)的图像复原方法。在生成器中加入ResNeXt残差块,提高模型的去模糊效果,并使用对抗损失和感知损失保证图像内容的一致性;以DiscoGAN为基础,构建2个GAN对非成对图像进行正向循环和反向循环,实现了模糊域到清晰域的相互转换。实验结果表明,该模型应用在GoPro数据集上结构相似性(SSIM)和峰值信噪比(PSNR)比其他同类去模糊模型平均提高了8.3%和15.1%,在桂林市交通数据集上测试展现出较好的迁移能力及泛化能力。

Due to weather changes and vehicle movement,the images captured by the intelligent monitoring system are blurred.Aiming at the problems of blurred traffic images and low resolution caused by light and vehicle movement,an image restoration method based on Generative Adversarial Network is proposed.Add the ResNeXt block to the generator to improve the deblurring effect of the model,and use the confrontation loss and the perceptual loss to ensure the consistency of the image content.Based on DiscoGAN,two GANs are constructed to carry out the forward loop and the unpaired image The reverse cycle realizes the mutual conversion from the fuzzy domain to the clear domain.The experimental results show that the structural similarity(SSIM)and peak signal-to-noise ratio(PSNR)data of this model applied to the GoPro data set are improved by an average of 8.3%and 15.1%compared with other similar deblurring models.The test shows on the Guilin city traffic data set.A better migration ability and generalization ability have been developed to provide practical significance and technical support for the intelligent transportation system.