摘要
医学图像在重建过程中总会受到噪声干扰,对于此问题,本文提出了一种基于条件生成对抗网络(CGAN)的去噪方法,算法以完整图像作为网络的输入及输出,使生成的图像信息更加稳定可靠。为了适应CT图像的特点,本文对CGAN结构进行了改进,使其能够适应不同噪声水平下的加性高斯白噪声,为了提高效率,在判别器进行训练时采用了损失判别,且在Tensorflow环境下训练网络模型。实验结果表明,与其他传统图像去噪算法相比,本方法能在保留特征信息的同时有效减少图像中的噪声。
Medical images are always disturbed by noise in the reconstruction process.For this problem,a method of condition generation against network(CGAN)is proposed.The algorithm takes the complete image as the input and output of the network,which makes the generated image information more stable and reliable.In order to adapt to the characteristics of CT image,CGAN structure is improved to adapt to additive Gaussian white noise at different noise levels.In order to improve efficiency,loss discrimination is used in discriminator training,and the network model is trained in tensorflow environment.The experiment results show that compared with other traditional image denoising algorithms,this method can save the feature information and reduce the noise in the image more effectively.
作者
雷肖雪
孔慧华
陈昱同
潘晋孝
李雨
Lei Xiaoxue;Kong Huihua;Chen Yutong;Pan Jinxiao;Li Yu(School of Science,North China University,Taiyuan 030051,China;Shanxi Key Laboratory of Signal Capturing&Processing,North University of China)
出处
《单片机与嵌入式系统应用》
2021年第3期37-41,共5页
Microcontrollers & Embedded Systems
基金
国家自然科学基金项目(61971381,61871351)
山西省自然科学基金项目(201701D221121)
信息探测与处理山西省重点实验室基金项目(ISPT2020-3)。
关键词
图像去噪
条件生成对抗网络
图像重建
深度学习
image denoising
conditional generation countermeasure network
image reconstruction
deep learning