目的图像修复技术虽然取得了长足进步,但是当图像中缺失区域较大时,非缺失区域提供的信息量非常有限,从而导致难以产生语义信息一致的内容来增强修复图像和真实图像的视觉一致性;同时图像修复常使用两阶段网络结构,基于该结构的模型不...目的图像修复技术虽然取得了长足进步,但是当图像中缺失区域较大时,非缺失区域提供的信息量非常有限,从而导致难以产生语义信息一致的内容来增强修复图像和真实图像的视觉一致性;同时图像修复常使用两阶段网络结构,基于该结构的模型不仅需要较长的训练时间,还会导致图像修复效果对第1阶段输出结果依赖性较强。针对上述问题,提出了一种基于双解码器的增强语义一致的图像修复方法。方法使用双解码器网络结构消除两阶段修复方法中存在的依赖性问题,同时有效缩短模型的训练时间;利用一致性损失、感知损失和风格损失,更好地捕获图像的上下文语义信息,解决图像修复任务中出现的视觉不一致的问题。此外,本文使用了跳跃连接,并引入多尺度注意力模块和扩张卷积,进一步提高了网络的特征提取能力。结果为了公正地评价,在CelebA、Stanford Cars和UCF Google Street View共3个数据集上对具有规则和不规则缺失区域的图像分别进行实验,采用客观评价指标:均方误差(L_(2))、峰值信噪比(peak signal-to-noise ratio,PSNR)、结构相似性(structural similarity,SSIM)、FID(Fréchet inception distance)和IS(inception score)进行评价。实验结果表明本文方法修复的图像不仅在视觉上有明显的提升,而且取得了较优的数值。如规则缺失区域下,在CelebA数据集中,本文方法的FID(越小越好)比性能第2的模型在数值上减少了39.2%;在UCF Google Street View数据集中,本文方法的PSNR比其他模型在数值上分别提高了12.64%、6.77%、4.41%。结论本文方法有效减少了模型的训练时间,同时消除了两阶段网络模型中的依赖性问题,修复的图像也呈现出更好的视觉一致性。展开更多
Soft decode-and-forward(DF) can combine the advantages of both amplify-and-forward and hard DF in relay channels. In this paper, we propose a low-complexity soft DF scheme based on polar codes, which features two key ...Soft decode-and-forward(DF) can combine the advantages of both amplify-and-forward and hard DF in relay channels. In this paper, we propose a low-complexity soft DF scheme based on polar codes, which features two key techniques: a low-complexity cyclic redundancy check(CRC) aided list successive cancellation(CALSC) decoder and a soft information calculation method. At the relay node, a low-complexity CALSC decoder is designed to reduce the computational complexity by adjusting the list size according to the reliabilities of decoded bits. Based on the path probability metric of the CALSC decoder, we propose a method to compute the soft information of the decoded bits in CALSC. Simulation results show that our proposed scheme outperforms the soft DF based on low-density parity-check codes and the soft DF with belief propagation or soft cancellation decoder, especially in the case when the source-relay channel is at the high signal-to-ratio region.展开更多
文摘目的图像修复技术虽然取得了长足进步,但是当图像中缺失区域较大时,非缺失区域提供的信息量非常有限,从而导致难以产生语义信息一致的内容来增强修复图像和真实图像的视觉一致性;同时图像修复常使用两阶段网络结构,基于该结构的模型不仅需要较长的训练时间,还会导致图像修复效果对第1阶段输出结果依赖性较强。针对上述问题,提出了一种基于双解码器的增强语义一致的图像修复方法。方法使用双解码器网络结构消除两阶段修复方法中存在的依赖性问题,同时有效缩短模型的训练时间;利用一致性损失、感知损失和风格损失,更好地捕获图像的上下文语义信息,解决图像修复任务中出现的视觉不一致的问题。此外,本文使用了跳跃连接,并引入多尺度注意力模块和扩张卷积,进一步提高了网络的特征提取能力。结果为了公正地评价,在CelebA、Stanford Cars和UCF Google Street View共3个数据集上对具有规则和不规则缺失区域的图像分别进行实验,采用客观评价指标:均方误差(L_(2))、峰值信噪比(peak signal-to-noise ratio,PSNR)、结构相似性(structural similarity,SSIM)、FID(Fréchet inception distance)和IS(inception score)进行评价。实验结果表明本文方法修复的图像不仅在视觉上有明显的提升,而且取得了较优的数值。如规则缺失区域下,在CelebA数据集中,本文方法的FID(越小越好)比性能第2的模型在数值上减少了39.2%;在UCF Google Street View数据集中,本文方法的PSNR比其他模型在数值上分别提高了12.64%、6.77%、4.41%。结论本文方法有效减少了模型的训练时间,同时消除了两阶段网络模型中的依赖性问题,修复的图像也呈现出更好的视觉一致性。
基金supported by the National Natural Science Foundation of China(No.61171099,No.61671080),Nokia Beijing Bell lab
文摘Soft decode-and-forward(DF) can combine the advantages of both amplify-and-forward and hard DF in relay channels. In this paper, we propose a low-complexity soft DF scheme based on polar codes, which features two key techniques: a low-complexity cyclic redundancy check(CRC) aided list successive cancellation(CALSC) decoder and a soft information calculation method. At the relay node, a low-complexity CALSC decoder is designed to reduce the computational complexity by adjusting the list size according to the reliabilities of decoded bits. Based on the path probability metric of the CALSC decoder, we propose a method to compute the soft information of the decoded bits in CALSC. Simulation results show that our proposed scheme outperforms the soft DF based on low-density parity-check codes and the soft DF with belief propagation or soft cancellation decoder, especially in the case when the source-relay channel is at the high signal-to-ratio region.
