摘要
针对传统图像超分辨重建技术中存在的特征丢失和缺乏高频细节的问题,在生成对抗网络的基础上结合注意力机制对网络进行改进.生成网络中通过多尺度残差注意力模块,学习不同尺度的图像特征,增强对图像高频细节的学习;再通过整体注意力模块,进一步捕获更多的信息特征,提高网络对图像细节的还原能力,用于最终重建.判别网络中使用非对称卷积替代传统卷积,减少参数计算量;并引入自注意力机制更精确地获取图像全局信息,提高网络重建性能.实验结果表明,重建后图像和原始图像相比具有更多的高频纹理细节,与7种常见的图像超分辨方法相比,PSNR(Picture Signal to Noise Ratio)平均提升约2.43 dB,SSIM(Structural Similarity Image Measurement)平均提升约0.1.
In response to the problems of feature loss and lack of high-frequency details in traditional image super-resolution reconstruction techniques,the network is improved by combining attention mechanism with the generation of adversarial networks.In the generation network,multi-scale residual attention modules are used to learn image features of different scales,enhance the learning of high-frequency details of the image;and then the entirety attention module is used to further capture more information features,improve the network′s ability to restore image details,and use it for final reconstruction.Using asymmetric convolution instead of traditional convolution in discriminative networks reduces parameter computation and introduces a self attention mechanism to more accurately obtain global image information,improving network reconstruction performance.The experimental results show that the reconstructed image has more high-frequency texture details compared to the original image.Compared with the seven common image super-resolution methods,the PSNR(Picture Signal to Noise Ratio)average improvement is about 2.43 dB,and the SSIM(Structural Similarity Image Measurement)average improvement is about 0.1.
作者
杨云
杨欣悦
张小璇
YANG Yun;YANG Xin-yue;ZHANG Xiao-xuan(School of Electronic Information and Artificial Intelligence,Shaanxi University of Science&Technology,Xi′an 710021,China)
出处
《陕西科技大学学报》
北大核心
2024年第2期216-223,232,共9页
Journal of Shaanxi University of Science & Technology
基金
国家自然科学基金项目(61971272,61601271)
国家重点研发计划项目(2019YFC1520204)。
关键词
生成对抗网络
多尺度残差融合
注意力机制
generative adversarial network
multi-scale residual fusion
attention mechanism
