摘要
Abstract
Government data contains a large amount of image data,which plays a crucial role in recording key information.The vagueness often appears in government images,which brings great trouble to the extraction and utilization of information.On this basis,a generative adversarial network(GAN)based blurry image restoration algorithm(GovRGAN)is proposed.In this algorithm,the GAN is used for the image restoration,which can effectively learn and recover detailed image information.It is composed of generator and discriminator.The generator of the GAN is pre-trained by means of the trained weight parameters of U-Net network.The convolutional neural network is used as the discriminator to distinguish between real images and those generated by the generator.In order to validate the algorithm's effectiveness,a government dataset with 1 500 invoice vouchers is constructed,aiming to provide sufficient and diverse training samples for the model.The motion blur and defocus blur are used for the degradation processing,making the data closer to blurry images in reality.The comparative experiments were conducted on this dataset between GovRGAN,AutoEncoder network,and U-Net,verifying that GovRGAN exhibits excellent performance in restoring blurred government images,and the quality of the restored images has been improved significantly.On the motion fuzzy dataset,in comparison with the U-Net network,the PSNR and SSIM values of the proposed algorithm are improved by 9.664 dB and 0.157,respectively.关键词
政务数据处理/模糊图像复原/生成对抗网络/卷积神经网络/AutoEncoder网络/U-Net网络Key words
government data processing/blurry image restoration/generative adversarial network/convolutional neural network/AutoEncoder network/U-Net network分类
信息技术与安全科学
