摘要
该方法是将低分辨率图像直接作为输入,逐级预测金字塔层的残差图像,选择特定的初始化方法对网络权值进行初始化,加快模型收敛,并引入多通道映射提取更加丰富的特征,采用卷积级联,共享权重的方式进行图像超分辨率重构,改进的模型可以更好地重建出图像的纹理和细节。
In order to solve these problems,an image super-resolution algorithm(LapMSRN)based on Laplacian pyramid structure for multichannel convolution network is proposed.This approach is the low resolution images directly as input,step by step to predict residual image pyramid layer,select a specific initialization method initialized weights of the network,to speed up the model convergence,the characteristics of the introduction of multichannel mapping to extract more rich,using convolution cascade,share the weight of image super-resolution reconstruction in the form of the proposed improved model can better reconstruction of the image texture and details.
作者
张倩宇
ZHANG Qian-yu(School of Mathematics and Computer Science,Shanxi Normal University, Linfen Shanxi 041000)
出处
《数字技术与应用》
2018年第10期71-72,共2页
Digital Technology & Application
关键词
图像超分辨率重建
拉普拉斯金字塔
残差
多通道
卷积级联
image super-resolution reconstruction
the Laplace pyramid
residual
the multichannel
convolution cascade
