电子学报 ›› 2019, Vol. 47 ›› Issue (12): 2622-2629.DOI: 10.3969/j.issn.0372-2112.2019.12.023

• 学术论文 • 上一篇    下一篇

一种非开关型快速随机脉冲噪声降噪算法

徐少平, 刘婷云, 罗洁, 张贵珍, 李崇禧   

  1. 南昌大学信息工程学院, 江西南昌 330031
  • 收稿日期:2018-10-08 修回日期:2019-08-13 出版日期:2019-12-25
    • 通讯作者:
    • 罗洁
    • 作者简介:
    • 徐少平 男,1976年5月出生于江西省九江市.博士,南昌大学信息工程学院计算机科学与技术系教授,博士生导师.主要研究方向为图形图像处理、机器视觉、虚拟手术仿真等.E-mail:xushaoping@ncu.edu.cn;刘婷云 女,1996年10月出生于江西省抚州市.现为南昌大学硕士研究生,主要研究方向为图像处理与计算机视觉.E-mail:416114517210@email.ncu.edu.cn;张贵珍 女,1993年5月出生于江西省赣州市.现为南昌大学硕士研究生,主要研究方向为图像处理与计算机视觉.E-mail:406130917331@email.ncu.edu.cn;李崇禧 男,1994年8月出生于江西省吉安市.现为南昌大学硕士研究生,主要研究方向为图像处理与计算机视觉.E-mail:406130917315@email.ncu.edu.cn
    • 基金资助:
    • 国家自然科学基金 (No.61662044,No.61163023,No.51765042); 江西省自然科学基金 (No.20171BAB202017)

A Fast Non-switching Random-Valued Impulse Noise Denoising Algorithm

XU Shao-ping, LIU Ting-yun, LUO Jie, ZHANG Gui-zhen, LI Chong-xi   

  1. School of Information Engineering, Nanchang University, Nanchang, Jiangxi 330031, China
  • Received:2018-10-08 Revised:2019-08-13 Online:2019-12-25 Published:2019-12-25
    • Corresponding author:
    • LUO Jie
    • Supported by:
    • National Natural Science Foundation of China (No.61662044, No.61163023, No.51765042); Natural Science Foundation of Jiangxi Province,  China (No.20171BAB202017)

摘要: 为提高现有开关型随机脉冲噪声(Random-Valued Impulse Noise,RVIN)降噪算法的降噪性能,提出了一种基于卷积神经网络的非开关型RVIN快速降噪算法(Fast Non-switching RVIN Denoising Algorithm,FNRDA).首先,利用噪声检测器随机地检测给定噪声图像中少量不同位置处的像素点;然后,将检测为RVIN噪声点的个数除以被检像素点总数转化为噪声比例值;最后,根据噪声比例值调用相应预先训练好的非开关型卷积神经网络降噪模型,快速且高质量地完成图像降噪任务.实验结果表明:所提出的非开关型FNRDA算法在各噪声比例下的综合性能(降噪效果和执行效率)优于经典的开关型RVIN降噪算法,适用于图像恢复、信号检测、无线通讯等实时系统中.

 

关键词: 降噪, 随机脉冲噪声, 非开关型, 卷积神经网络, 噪声比例值, 执行效率

Abstract: To improve denoising effect and execution efficiency of the existing switching random-valued impulse noise (RVIN) removal algorithms, we propose a convolutional neural network (CNN)-based fast non-switching RVIN denoising algorithm (FNRDA), which consists of two serial CNN-based modules, i.e., noise detector and denoiser. Specifically, we first use the noise detector to detect some randomly selected pixels of a given noisy image. Then we divide the number of the detected noisy pixels by the total number of detected pixels to convert it into noise ratio, which can be treated as a measure of the distortion level for the given noisy image. Finally, according to the estimated noise ratio, we exploit the corresponding pre-trained non-switching CNN-based denoising model to remove RVIN efficiently with high quality. Experimental results show that, the proposed non-switching RVIN removal algorithm outperforms the classical switching ones in terms of denoising effect and execution efficiency across various noise ratios. This advantage makes it more attractive and practical in the real-time applications such as image restoration, signal detection, wireless communication, etc.

 

Key words: denoising, random-valued impulse noise, non-switching, convolutional neural network (CNN), noise ratio, computational efficiency

中图分类号: