摘要
针对低分辨率图像到高分辨率图像的映射函数解空间极大,导致超分辨率重建模型难以产生细致纹理的问题,本文提出一种结合双注意力和结构相似度量的图像超分辨率重建网络。以改进的U-Net网络模型作为基本结构,引入针对低级别视觉任务的数据增强方法,增加样本多样性。编码器部分由卷积层和自适应参数线性整流激活函数(Dynamic ReLU)组成。同时引入了一种残差双注意力模块(Residual Dual Attention Module,RDAM)与像素重组(PixelShuffle)模块共同构成解码器,通过上采样操作,逐级放大图像。为了使生成图像更加符合人眼视觉特性,采用了一种结合结构相似度量准则的损失函数,增强网络约束。实验结果表明重建图像的质量对比SRCNN,在Set5、Set14、BSD100和Urban100标准测试集上的平均PSNR提升约1.64 dB,SSIM提升约0.047。本文方法能够使重建的图像纹理细节更丰富,有效地减少了映射函数可能的解空间。
Aiming at the problem that the solution space of mapping function from low resolution image to high resolution image is extremely large,which makes it difficult for super-resolution reconstruction models to generate detailed textures,this paper proposes a image super resolution that combines dual attention and structural similarity measure.With the improved U-Net network model as the basic structure,the data augmentation methods for low-level vision tasks are introduced to increase sample diversity.The encoder is composed of a convolution layer and an adaptive parameter linear rectifier function(Dynamic ReLU).At the same time,a residual dual attention module(RDAM)is introduced,which forms a decoder together with the PixelShuffle module.The image is enlarged gradually through the up-sampling operation.In order to make the generated image more in line with the human visual characteristics,a loss function combined with structural similarity measurement criteria is proposed to enhance the network constraints.The experimental results show that the average PSNR of the quality of the reconstructed image on the Set5,Set14,BSD100 and Urban100 standard test sets is improved by about 1.64 dB,and the SSIM is improved by about 0.047 compared with SRCNN.The proposed method can make the reconstructed image texture more detailed and reduce the possible solution space of the mapping function effectively.
作者
黄友文
唐欣
周斌
HUANG You-wen;TANG Xin;ZHOU Bin(School of Information Engineering, Jiangxi University of Science and Technology, Ganzhou 341000, China)
出处
《液晶与显示》
CAS
CSCD
北大核心
2022年第3期367-375,共9页
Chinese Journal of Liquid Crystals and Displays
基金
江西省教育厅科技项目(No.GJJ180443)。
关键词
超分辨率
U-Net网络
数据增强
双注意力
结构相似
super-resolution
U-net network
data augmentation
dual attention
structural similarity
