人工智能培训

搜索

机器学习论文:DEEP RESIDUAL AUTOENCODER用于高质量的独立JPEG恢复(Deep Residual Autoencoder for qua

[复制链接]
rickyc 发表于 2019-3-15 12:08:47 | 显示全部楼层 |阅读模式
rickyc 2019-3-15 12:08:47 805 0 显示全部楼层
机器学习论文:DEEP RESIDUAL AUTOENCODER用于高质量的独立JPEG恢复(Deep Residual Autoencoder for quality independent JPEG restoration)在本文中,我们提出了一种深度残差自动编码器,它利用剩余残差密度块(RRDB)去除JPEG压缩图像中的伪像,这些伪像独立于所使用的品质因子(QF)。所提出的方法利用了深度剩余网络的学习能力和JPEG压缩管道的先验知识。所提出的模型在YCbCr颜色空间中运行,并使用两个不同的自动编码器在两个阶段中执行JPEG伪像恢复:第一个恢复亮度通道二维卷积;第二个,使用恢复的亮度通道作为指导,恢复色度通道,消除3D卷积。三个广泛使用的基准数据集(即LIVE1,BDS500和CLASSIC-5)的广泛实验结果表明,我们的模型能够在所考虑的所有评估指标(即PSNR,PSNR-B和SSIM)方面超越现有技术水平。 。该结果是显着的,因为现有技术中的方法对于每种压缩质量使用不同的权重集,而所提出的模型对所有这些使用相同的权重,使得其适用于在用于压缩的QF未知的野外图像。此外,当应用于训练期间未见的压缩质量时,所提出的模型显示出比现有技术方法更强的鲁棒性。
In this paper we propose a deep residual autoencoder exploitingResidual-in-Residual Dense Blocks (RRDB) to remove artifacts in JPEG compressedimages that is independent from the Quality Factor (QF) used.The proposedapproach leverages both the learning capacity of deep residual networks andprior knowledge of the JPEG compression pipeline.The proposed model operatesin the YCbCr color space and performs JPEG artifact restoration in two phasesusing two different autoencoders: the first one restores the luma channelexploiting 2D convolutions;the second one, using the restored luma channel asa guide, restores the chroma channels explotining 3D convolutions.Extensiveexperimental results on three widely used benchmark datasets (ie LIVE1,BDS500, and CLASSIC-5) show that our model is able to outperform the state ofthe art with respect to all the evaluation metrics considered (ie PSNR,PSNR-B, and SSIM).This results is remarkable since the approaches in the stateof the art use a different set of weights for each compression quality, whilethe proposed model uses the same weights for all of them, making it applicableto images in the wild where the QF used for compression is unkwnown.Furthermore, the proposed model shows a greater robustness thanstate-of-the-art methods when applied to compression qualities not seen duringtraining.机器学习论文:DEEP RESIDUAL AUTOENCODER用于高质量的独立JPEG恢复(Deep Residual Autoencoder for quality independent JPEG restoration) VZfuRj33cf1y0HRs.jpg
URL地址:https://arxiv.org/abs/1903.06117     ----pdf下载地址:https://arxiv.org/pdf/1903.06117    ----机器学习论文:DEEP RESIDUAL AUTOENCODER用于高质量的独立JPEG恢复(Deep Residual Autoencoder for quality independent JPEG restoration)
回复

使用道具 举报

您需要登录后才可以回帖 登录 | 立即注册

本版积分规则 返回列表 发新帖

rickyc当前离线
新手上路

查看:805 | 回复:0

快速回复 返回顶部 返回列表