结合SENet的密集卷积生成对抗网络图像修复方法

Dense Convolution Generative Adversarial Network Image Inpainting Method With SENet

本文针对数据集较小或者图像结构相对复杂的较大面积缺失的图像修复问题,提出结合SENet的密集卷积生成对抗网络图像修复方法.首先,采用生成对抗网络的思想,生成器使用密集卷积块捕捉图像中缺失部分的语义信息再利用;其次,取消密集卷积块之间的过渡层,引入SENet注意力机制SE模块,获取特征重要程度,增强特征信息指导能力;再次,在编码器和解码器之间引入跳跃连接,减少由于下采样而造成的信息损失;最后,通过引入对抗损失、MSE损失、TV损失增强网络的稳定性.所提模型在CelebA数据集进行实验.结果表明,所提算法的修复结果在图像语义、峰值信噪比(PSNR)和结构相似度(SSIM)3个方面均具有不错成效.

In this paper,a dense convolution generative adversarial network image inpainting method with SENet is proposed to solve the problem of missing images with small data sets or relatively complex image structures. Firstly,using the idea of generative adversarial network,the generator uses dense convolutional blocks to capture the semantic information of the missing part of the image for reuse. Secondly,the transition layer between dense convolutional blocks is cancelled. SE module of attentional mechanism from SENet is introduced to obtain the importance of features and enhance the guiding ability of feature information. Thirdly,skip connection is introduced between encoder and decoder to reduce the information loss caused by downsampling. Finally,adversarial loss,MSE loss and TV loss are introduced to enhance the stability of network. The proposed model was tested on CelebA dataset. The results show that the proposed algorithm has good performance in image semantics, peak signal-to-noise ratio( PSNR) and structural similarity index(SSIM).