基于生成对抗网络的煤矿工人脸图像超分辨方法  

Super-resolution method of miners′face image based on generative adversarial network

作　　者：高志军 冯娇娇 Gao Zhijun;Feng Jiaojiao(School of Computer&Information Engineering,Heilongjiang University of Science&Technology,Harbin 150022,China)

机构地区：[1]黑龙江科技大学计算机与信息工程学院,哈尔滨150022

出　　处：《黑龙江科技大学学报》2022年第6期828-835,共8页Journal of Heilongjiang University of Science And Technology

基　　金：黑龙江省省属高等学校基本科研业务费项目(2019-KYYWF-11)。

摘　　要：为重建清晰的煤矿工人脸图像,提出一种基于改进生成对抗网络的煤矿工人脸图像超分辨率重建方法。通过引入删除批量归一化层的残差密集网络加深网络的深度,提取图像的特征信息,采用亚像素卷积层和渐进式采样完成矿工人脸图像的重建,保证重建图像颜色的均匀性,利用相对鉴别器模型引导生成器重建高质量、多细节的图像,实验对比改进前后模型的清晰度。结果表明,文中算法获得的图像峰值信噪比提升至28.3663 dB,结构相似性值提高至0.7685。改进后的算法可以获得更加清晰的矿工人脸图像,整体性能优于其他算法。This paper seeks to reconstruct a clear face image of coal miners,and proposes a super resolution image reconstruction method of coal miners′face image based on improved generation confrontation network.The study includes extracting the feature information of the image by introducing the residual dense network of deleting the batch normalization layer to deepen the depth of the network;using the sub-pixel convolution layer and progressive sampling to complete the reconstruction of the miners′face image,and ensuring the uniformity of the reconstructed image color;using the relative discriminator model to guide the generator to reconstruct high-quality and more detailed images;and improving the clarity of the model by comparing the experiments early or late.The results show that PSNR of the image obtained according to the improved algorithm increases up to 28.3663 dB,and the SSIM increases to 0.7685.The improved algorithm can obtain clearer miners′face images,and the overall performance is better than other algorithms.

关 键 词：超分辨率图像重建 生成对抗网络 残差密集网络 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]