文章快速检索 高级检索

1. 沈阳建筑大学 机械工程学院, 辽宁 沈阳 110168;
2. 沈阳建筑大学 高档石材数控加工装备与技术国家地方联合工程实验室, 辽宁 沈阳 110168

Filtering image impulse noise by using a PCNN image noise reduction technique
YAN Haipeng1, WU Yuhou2
1. School of Mechanical Engineering, Shenyang Jianzhu University, Shenyang 110168, China;
2. National-Local Joint Engineering Laboratory of High-Grade Stone Numerical Control Machining Equipments and Technology, Shenyang Jianzhu University, Shenyang 110168, China
Abstract: Traditional methods for image noise reduction typically damage the edges and details of an image, blur image contours, and thereby make them indistinct after image noise reduction is complete. To achieve better results in image noise reduction, we propose a pulse coupling neural network (PCNN) image noise reduction method based on a modified synaptic link strength and a modified threshold function. We simplified the basic PCNN model and adaptively changed the synaptic link strength value; further, we improved the threshold function by using a segmented attenuation function so as to improve the resolving power for different gray values of the given images. We improved the accuracy of our algorithm for identifying noise by positioning noise points according to the difference of firing times between the neuron and its surrounding neurons. Using this approach, we achieved better noise reduction results; our experimental results showed that our proposed method was able to accurately identify image impulse noise points and effectively remove these noise points. Further, through subjective evaluation, we observed that image edge details were also protected.
Key words: image noise reduction     pulse coupling neural network     synaptic link strength     threshold function     resolving power

1 PCNN神经元模型 1.1 PCNN介绍

PCNN是由许多神经元相互链接形成的一种动态非线性神经网络。如图 1所示，一个PCNN神经元主要由接收外部刺激信号部分、对信号进行调制部分以及根据信号强弱决定能否产生脉冲部分这3部分组成。

 图 1 基本PCNN神经元模型 Fig. 1 Basic PCNN neuron model

1.2 简化PCNN模型

 (1)
 (2)
 (3)
 (4)
 (5)

2 改进PCNN模型降噪 2.1 改进突触链接强度

 (6)

2.2 改进阈值函数

 (7)

 图 2 动态阈值衰减曲线 Fig. 2 Dynamic threshold attenuation curve
2.3 噪声定位

2.4 PCNN滤波算法

1) 初始化PCNN，设置各个参数初始值，同时令神经元均处于熄火状态，即Yij=0。

2) 输入含噪图像，在PCNN中按式 (1)~(4) 及式 (6) 和式 (7) 循环迭代，直到所有神经元均点火发放脉冲，同时记录神经元初次点火时间于点火时间矩阵T中。

3) 应用2.3小节介绍的方式定位噪声。对大值噪声点和小值噪声点均采用 (2m+1)×(2m+1) 窗口局部中值滤波。除以上情况外，直接输出原灰度值。

4) 输出降噪后图像。

3 实验验证与结果分析

 图 3 测试图像的原始图像 Fig. 3 The original images for test

 图 4 Lena图像添加密度10%的椒盐噪声降噪效果 Fig. 4 Noise reduction results for Lena image added the salt and pepper noise with a density of 10%
 图 5 Deer图像添加密度10%的椒盐噪声降噪效果 Fig. 5 Noise reduction results for Deer image added the salt and pepper noise with a density of 10%

 图 6 降噪细节比较 Fig. 6 Detail comparison of noise reduction results

 评价指标 Lena Deer PSNR MSE ISNR PSNR MSE ISNR 加噪图像 15.515 1826 0 15.048 2034 0 均值降噪 23.450 290.5 -8.04 20.813 539.3 -5.70 中值降噪 31.003 51.62 -15.57 23.520 289.2 -8.41 本文算法 38.265 9.70 -22.81 32.423 37.22 -17.32

 评价指标 6% 10% 14% 20% 30% 50% 加噪图像 17.55 15.52 13.98 12.41 10.71 8.45 均值降噪 25.25 23.54 22.18 22.08 20.28 18.36 中值降噪 31.83 31.00 30.21 29.57 27.55 24.77 文献[10] 34.14 33.90 — 32.42 30.15 — 文献[15] 33.77 32.75 31.95 30.51 — — 本文算法 40.97 38.28 36.57 34.64 31.62 27.49

 序号 K0 K1 K2 1 4 8 4 2 8 15 8 3 16 30 16 4 4 15 4 5 8 30 8

 序号 PSNR MSE ISNR TIME 1 37.998 4 10.327 2 -22.544 2 45.834 5 2 38.503 5 9.182 7 -23.050 3 30.266 5 3 36.071 0 16.076 8 -20.590 5 21.705 7 4 38.247 6 9.916 8 -22.770 0 40.646 4 5 37.925 3 10.486 7 -22.452 9 27.017 5

4 结束语

 [1] NAKARIYAKUL S. Fast spatial averaging: an efficient algorithm for 2D mean filtering[J]. The journal of supercomputing, 2013, 65(1): 262-273. DOI:10.1007/s11227-011-0638-9. [2] YUAN Xinxing, WEN Peng, FAN Xiuxiang, et al. A local pixel distribution based self-adaptive median filter for removal of pepper and salt noise[J]. IFAC proceedings volumes, 2013, 46(20): 63-67. DOI:10.3182/20130902-3-CN-3020.00179. [3] WANG Huiyan, ZHENG Jia. Comparative study of tongue image denoising methods[J]. Journal of computers, 2013, 8(3): 787-794. [4] 张文兴, 闫海鹏, 王建国. 基于改进脉冲耦合神经网络的数据降噪方法研究[J]. 机械设计与制造, 2015(2): 25-28. ZHANG Wenxing, YAN Haipeng, WANG Jianguo. Research on data noise reduction method based on modified PCNN[J]. Machinery design & manufacture, 2015(2): 25-28. [5] WANG Zhaobin, MA Yide, CHENG Feiyan, et al. Review of pulse-coupled neural networks[J]. Image and vision computing, 2010, 28(1): 5-13. DOI:10.1016/j.imavis.2009.06.007. [6] SUBASHINI M M, SAHOO S K. Pulse coupled neural networks and its applications[J]. Expert systems with applications, 2014, 41(8): 3965-3974. DOI:10.1016/j.eswa.2013.12.027. [7] 沈艳, 张晓明, 韩凯歌, 等. PCNN图像分割技术研究[J]. 现代电子技术, 2014, 37(2): 38-41. SHEN Yan, ZHANG Xiaoming, HAN Kaige, et al. Research of image segmentation technology based on PCNN[J]. Modern electronics technique, 2014, 37(2): 38-41. [8] 周东国, 高潮, 郭永彩. 一种参数自适应的简化PCNN图像分割方法[J]. 自动化学报, 2014, 40(6): 1191-1197. ZHOU Dongguo, GAO Chao, GUO Yongcai. Adaptive simplified PCNN parameter setting for image segmentation[J]. Acta automatica sinica, 2014, 40(6): 1191-1197. [9] 李翔. 基于脉冲耦合神经网络的图像识别和图像检索算法研究[D]. 昆明: 云南大学, 2014. LI Xiang. Research on image recognition and image retrieval algorithm based on pulse coupled neural network[D]. Kunming: Yunnan University, 2014. [10] 张文兴, 闫海鹏, 王建国. 一种基于脉冲耦合神经网络的图像降噪方法[J]. 图学学报, 2015, 36(1): 47-51. ZHANG Wenxing, YAN Haipeng, WANG Jianguo. A method for image de-noising based on pulse coupled neural network[J]. Journal of graphics, 2015, 36(1): 47-51. [11] 李海燕, 张榆锋, 施心陵, 等. 基于脉冲耦合神经网络的自适应图像滤波[J]. 计算机应用, 2011, 31(4): 1037-1039. LI Haiyan, ZHANG Yufeng, SHI Xinling, et al. Adaptive filtering method for images based on pulse-coupled neural network[J]. Journal of computer applications, 2011, 31(4): 1037-1039. [12] 张艳珠, 李媛, 李小娟. 简化型PCNN的混合噪声图像滤波算法研究[J]. 控制工程, 2013, 20(5): 829-832. ZHANG Yanzhu, LI Yuan, LI Xiaojuan. The research of hybrid noise filtering for images based on pulse coupled neural network[J]. Control engineering of China, 2013, 20(5): 829-832. [13] 刘勍. 基于脉冲耦合神经网络的图像处理若干问题研究[D]. 西安: 西安电子科技大学, 2011. LIU Qing. Research on several issues about image processing based on pulse coupled neural networks[D]. Xi'an: Xidian University, 2011. [14] 樊洪斌. 脉冲耦合神经网络在医学图像处理中的应用研究[D]. 桂林: 广西师范大学, 2009. FAN Hongbin. The applications in the medical image processing based on pulse coupled neural network[D]. Guilin: Guangxi Normal University, 2009. [15] 刘勍, 马义德. 一种基于PCNN赋时矩阵的图像去噪新算法[J]. 电子与信息学报, 2008, 30(8): 1869-1873. LIU Qing, MA Yide. A new algorithm for noise reducing of image based on PCNN time matrix[J]. Journal of electronics & information technology, 2008, 30(8): 1869-1873.
DOI: 10.11992/tis.201605027

0

#### 文章信息

YAN Haipeng, WU Yuhou

Filtering image impulse noise by using a PCNN image noise reduction technique

CAAI Transactions on Intelligent Systems, 2017, 12(2): 272-278
http://dx.doi.org/10.11992/tis.201605027