双通道单图像超分辨率卷积神经网络

Dual path convolutional neural network for single image super-resolution

基于残差与稠密结构的卷积神经网络的单张图像的超分辨率(SISR)方法显著地提升了重建的性能,然而残差网络侧重于特征的复用,而稠密连接可以实现对新特征的探索,为了结合两者的优势,本文设计了一种双通道图像超分辨率深度卷积神经网络,将特征映射在第三维度上分割成两条路径,一条以残差的形式进行连接,另一条以稠密的方式进行跳跃连接。同时,在网络末端引入解卷积层来放大特征映射,显著加速了计算,并且直接从低分辨率图像到高分辨率图像之间进行端到端的映射。评估结果表明,本文方案取得了比当前绝大多数网络模型更高的峰值信噪比(PSNR)。

Recent researches have shown that deep convolutional neural networks can significantly boost the performance of Single-image super-resolution(SISR).In particular,residual network and densely convolutional network can improve performance remarkably.Both path topologies are proposed to alleviate the vanishing-gradient problem of deep convolution networks.Since the residual network enables feature re-usage and the dense skip connections enables new features exploration.A dual path network is proposed for single-image super-resolution by combining the residual network and the dense skip connections in a very deep network.In the proposed network,the feature maps are split into two paths,one path is propagated in the form of residual,and another path is propagated by dense skip connections.In addition,the deconvolution layers are integrated into the network to upscale the feature map which can significantly speedup the network,and the mapping is learn from the low-resolution image to the high-resolution image directly.The network is evaluated with four benchmark datasets.The simulation results demonstrate that the proposed network has much higher peak signal-to-noise ratio(PSNR),in contrast to most conventional state-of-art methods.