基于FPGA的PCNN图像噪声滤波器设计
|
摘要
随着嵌入式系统的发展,基于硬件实现的图像处理技术需求快速增长。软硬件可裁剪是嵌入式系统的特点,高速、实时性是硬件实现的优点,FPGA具有灵活的可编程和可重配置性能,FPGA器件的高集成度和高速处理能力使得用硬件实现图像处理算法成为可能。近年来基于FPGA的图像处理研究成为图像处理领域一个重要的研究方向。
基于PCNN和中值滤波器的图像脉冲噪声滤波器算法是本实验室多年来进行图像处理技术研究取得的成果之一。相比Matlab软件实现,将该脉冲噪声滤波器算法移植到FPGA硬件平台实现,具有速度快,性能好等的优点。论文主要进行PCNN脉冲噪声滤波器在FPGA平台上的实现研究。
论文设计的基于FPGA的PCNN脉冲噪声滤波器主要包括PCNN噪声检测模块、中值滤波模块、图像显示输出模块和Flash读写控制模块等。在实现过程中解决了以下几个问题：
(1)将PCNN结合中值滤波设计的脉冲噪声滤波器在硬件FPGA上实现,主要用PCNN的分类属性来判断噪声点,再用中值滤波来滤除噪声点。
(2)在FPGA上实现PCNN脉冲噪声检测时,提出对于高亮度脉冲噪声(盐噪声),使用较高的初始阈值,经过很少的点火次数(1至2次)即可定位出噪声点；而对于低亮度脉冲噪声(椒噪声),只需将图像进行反转,使低亮度变为高亮度,用同样的方法处理即可,这样使迭代次数大为减少。
(3)将Flash读写控制和VGA显示控制结合,VGA显示控制模块与滤波器输出连接,使滤波器输出直观在显示器上显示,同时将结果以文件形式保存,便于后期分析处理。
With the development of embedded systems, the requirement of hardware-based image processing grows rapidly. Embedded systems have a flexible feature in cutting hardware and software. And hardware implementation possesses the advantages of high-speed and real-time. It is possible that image processing algorithms are implemented in hardware, because FPGA has flexible programmable, reconfigurable performance, high integration and high-speed processing capability. In recent years, FPGA-based image processing becomes an important research direction in the field of image processing.
Image filter algorithms for impulse noise based on PCNN and the median filter has been one of the jobs in our laboratory for many years. Compared to the implementation of Matlab software, the implementation of FPGA hardware platform possesses high-speed and good performance and other advantages. In this paper, the PCNN impulse noise filter is implemented on the FPGA platform.
In this paper, the design of the filter mainly includes PCNN noise detect module, median filtering module, image display module and control module for read and write(in Flash). The following issues are resolved during the process of FPGA hardware implementation of the filter.
(1) The filter based on PCNN and the median filter for impulse noise is implemented in FPGA hardware platform. The noise points are mainly detected by the classification property of PCNN. Then the noise points are filtered by the median filter.
(2) When the impulse noises are detected in FPGA, the noises for high- intensity (salt noise) is processed by using a higher initial threshold value, and the noises can be located after firing a few times (1-2 times); while the noises for low-intensity (pepper noise) is detected, only converting low-intensity to high- intensity, the method of processing is the same. So, the number of iterations is largely decreased.
(3) It combines the Flash controller for reading and writing with the VGA display controller. And the output of the filter is connected to VGA display output module, so the output results can be displayed on the monitor. At the same time, the results are saved in the flash memory,So that they will be analysed and processed later.
基于PCNN和中值滤波器的图像脉冲噪声滤波器算法是本实验室多年来进行图像处理技术研究取得的成果之一。相比Matlab软件实现,将该脉冲噪声滤波器算法移植到FPGA硬件平台实现,具有速度快,性能好等的优点。论文主要进行PCNN脉冲噪声滤波器在FPGA平台上的实现研究。
论文设计的基于FPGA的PCNN脉冲噪声滤波器主要包括PCNN噪声检测模块、中值滤波模块、图像显示输出模块和Flash读写控制模块等。在实现过程中解决了以下几个问题：
(1)将PCNN结合中值滤波设计的脉冲噪声滤波器在硬件FPGA上实现,主要用PCNN的分类属性来判断噪声点,再用中值滤波来滤除噪声点。
(2)在FPGA上实现PCNN脉冲噪声检测时,提出对于高亮度脉冲噪声(盐噪声),使用较高的初始阈值,经过很少的点火次数(1至2次)即可定位出噪声点；而对于低亮度脉冲噪声(椒噪声),只需将图像进行反转,使低亮度变为高亮度,用同样的方法处理即可,这样使迭代次数大为减少。
(3)将Flash读写控制和VGA显示控制结合,VGA显示控制模块与滤波器输出连接,使滤波器输出直观在显示器上显示,同时将结果以文件形式保存,便于后期分析处理。
With the development of embedded systems, the requirement of hardware-based image processing grows rapidly. Embedded systems have a flexible feature in cutting hardware and software. And hardware implementation possesses the advantages of high-speed and real-time. It is possible that image processing algorithms are implemented in hardware, because FPGA has flexible programmable, reconfigurable performance, high integration and high-speed processing capability. In recent years, FPGA-based image processing becomes an important research direction in the field of image processing.
Image filter algorithms for impulse noise based on PCNN and the median filter has been one of the jobs in our laboratory for many years. Compared to the implementation of Matlab software, the implementation of FPGA hardware platform possesses high-speed and good performance and other advantages. In this paper, the PCNN impulse noise filter is implemented on the FPGA platform.
In this paper, the design of the filter mainly includes PCNN noise detect module, median filtering module, image display module and control module for read and write(in Flash). The following issues are resolved during the process of FPGA hardware implementation of the filter.
(1) The filter based on PCNN and the median filter for impulse noise is implemented in FPGA hardware platform. The noise points are mainly detected by the classification property of PCNN. Then the noise points are filtered by the median filter.
(2) When the impulse noises are detected in FPGA, the noises for high- intensity (salt noise) is processed by using a higher initial threshold value, and the noises can be located after firing a few times (1-2 times); while the noises for low-intensity (pepper noise) is detected, only converting low-intensity to high- intensity, the method of processing is the same. So, the number of iterations is largely decreased.
(3) It combines the Flash controller for reading and writing with the VGA display controller. And the output of the filter is connected to VGA display output module, so the output results can be displayed on the monitor. At the same time, the results are saved in the flash memory,So that they will be analysed and processed later.
引文
[1]张祥光.非线性理论在图像抑噪中的应用研究[硕士论文].兰州：兰州大学.2002.
[2]马义德,史飞,李廉,安黎哲.一种基于脉冲耦合神经网络的脉冲噪声滤波器设计[J].生物医学工程学杂志.2004,(06)：1019-1023.
[3]石美红,张军英,朱欣娟,张晓滨.基于PCNN的图像高斯噪声滤波的方法[J].计算机应用.2002,22(6)：1-4.
[4]Eckhorn R, Reitboeck H J, Arndt M,Dicke P W. Feature linking via synchronization among distributed assemblies:Simulation of results from cat cortex [J]. Neural Comput. 1990,2(3):293-307.
[5]马义德,戴若兰,李廉.一种基于脉冲耦合神经网络和图像熵的自动图像分割方法[J].通信学报.2002,(01)：46-51.
[6]Johnson J L,Ritter D. Observation of periodic waves in a Pulse-coupled neural network[J]. Optics Letters.1993,18(15):1253-1255.
[7]Johnson J L. Pulse-coupled neural nets:Translation, rotation, scale, distortion, and intensity signal invariance for images[J]. Applied Optics.1994,33(26):6239-6253.
[8]Kuntimad G, Ranganath H S. Perfect image segmentation using pulse coupled neural networks[J]. IEEE Transactions on Neural Networks.1999,10(3):591-598.
[9]Johnson J L, Padgett M L. PCNN models and applications[J]. IEEE Trans. on Neural Networks.1999,10 (3):480-498.
[10]马义德.生物信息的数字图像处理技术研究[博士论文].兰州：兰州大学.2001.
[11]张军英,卢志军,石林,董继扬,石美红.基于脉冲耦合神经网络的椒盐噪声图像滤波[J].中国科学E辑.2004,34(8)：882-894.
[12]刘勃,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,(08)：1869-1873.
[13]卢桂馥,王勇,窦易文.一种新的基于PCNN的图像脉冲噪声滤波算法[J].计算机技术与发展.2007,17(12)：83-85.
[14]马义德,张红娟PCNN与灰度形态学相结合的图像去噪方法[J].北京邮电大学学报.2008,31(02)：108-112.
[15]王巍,黎明,刘高航.基于脉冲耦合神经网络的彩色图像滤波新方法[J].计算机工程与设计.2007,28(14)：3413-3415.
[16]陈锐,马义德,郑晓,袁树林.基于PCNN和视觉特性的图像对比度增强[J］.微计算机信息.2010,26(4-1)：108-111.
[17]徐光柱,刘鸣,任东,马义德,刘晓丽.基于脉冲耦合神经网络的多区域图像分割[J].山东大学学报(理学版).2010,45(07)：86-93.
[18]张北斗,马义德,林冬梅,张立文.基于脉冲耦合神经网络的超声图像病理区域检测算法[J].兰州大学学报(自然科学版).2008,44(01)：61-64.
[19]绽琨,张红娟,马义德,刘丽,田乐.交叉皮层模型及其在图像处理中的应用[J].北京邮电大学学报.2009,32(04)：40-45.
[20]李志强,程飞燕,安黎哲,马义德,高清祥.基于多维PCNN的彩色枸杞细胞显微图像 的边缘检测[J].兰州大学学报(自然科学版).2009,45(05)130-134.
[21]刘映杰,马若飞,朱望飞,者吴,绽琨,马义德.基于多阈值PCNN的运动目标检测算法[J].计算机应用.2009,29(03)：739-741.
[22]朱望飞,马义德,邱秀清.基于PCNN的高斯混合模型运动检测改进方法[J].兰州大学学报(自然科学版).2009,45(02)：129-131.
[23]王新PCNN混沌特性与硬件实现研究[硕士论文].兰州：兰州大学.2010.
[24]赵釜.基于Cyclone Ⅱ系列FPGA的图像实时采集与预处理系统研究[硕士论文].重庆：重庆大学.2009.
[25]张志刚FPGA与SOPC设计教程——DE2实践[M].西安：西安电子科技大学出版社.2007.4：9-30.
[26]仙云森.基于FPGA的图像处理算法研究及硬件设计[硕士论文].大连：大连理工大学.2008.
[27]Altera Corporation. Quartus Ⅱ Handbook [EB/OL]. http://www.altera.com.
[28]Terasic Technologies Inc. DE2-70 User manual [EB/OL]. http://www.terasic.com.cn.
[29]Altera Corporation. Altera DE2-70 Development and Education Board User Manual [EB/OL]. http://www.altera.com.
[30]Altera Corporation. De2-70 Handbook [EB/OL]. http://www.altera.com.
[31]史飞.脉冲耦合神经网络在图像处理中的应用研究[硕士论文].兰州：兰州大学.2003.
[32]李洋波,赵不贿.基于FPGA的图像中值滤波器的硬件实现[J].现代电子技术.2008,(22)：99-101.
[33]马义德,李廉,绽锟,王兆斌.脉冲耦合神经网络与数字图像处理[M].北京：科学出版社.2008.7.
[34]付昱强.基于FPGA的图像处理算法的研究与硬件设计[硕十论文].南昌：南昌大学.2006.
[35]刘健鹏,陈卫东,钱钧.基于FPGA的实时中值滤波器设计[J].应用光学.2007,28(06)：712-715.
[36]郑鹤,王鲁平,李飙.一种快速的二维中值滤波算法及其硬件实现[J].信息与电子工程.2005,3(04)：245-248.
[37]李轶博,李小兵,周娴.基于FPGA的快速中值滤波器设计与实现[J].液晶与显示.2010,25(02)：292-296.
[38]李飞飞,刘伟宁,王艳华.改进的中值滤波算法及其FPGA快速实现[J].计算机工程.2009,35(14)：175-177.
[39]陈加成,徐熙平,吴琼.基于FPGA的中值滤波算法研究与硬件设计[J].长春理[大学学报(自然科学版).2008,31(01)8-10.
[40]黄艳军.基于FPGA的数字图像预处理算法研究[硕十论文].南京：南京理工大学.2009.
[41]徐大鹏,李从善.基于FPGA的数字图像中值滤波器设计[J].电子器件.2006,29(04)1114-1117.
[42]胡越黎,计慧杰,吴频,宣祥光.图像的中值滤波算法及其FPGA实现[J].计算机测量与控制.2008,16(11)：1672-1675.
[2]马义德,史飞,李廉,安黎哲.一种基于脉冲耦合神经网络的脉冲噪声滤波器设计[J].生物医学工程学杂志.2004,(06)：1019-1023.
[3]石美红,张军英,朱欣娟,张晓滨.基于PCNN的图像高斯噪声滤波的方法[J].计算机应用.2002,22(6)：1-4.
[4]Eckhorn R, Reitboeck H J, Arndt M,Dicke P W. Feature linking via synchronization among distributed assemblies:Simulation of results from cat cortex [J]. Neural Comput. 1990,2(3):293-307.
[5]马义德,戴若兰,李廉.一种基于脉冲耦合神经网络和图像熵的自动图像分割方法[J].通信学报.2002,(01)：46-51.
[6]Johnson J L,Ritter D. Observation of periodic waves in a Pulse-coupled neural network[J]. Optics Letters.1993,18(15):1253-1255.
[7]Johnson J L. Pulse-coupled neural nets:Translation, rotation, scale, distortion, and intensity signal invariance for images[J]. Applied Optics.1994,33(26):6239-6253.
[8]Kuntimad G, Ranganath H S. Perfect image segmentation using pulse coupled neural networks[J]. IEEE Transactions on Neural Networks.1999,10(3):591-598.
[9]Johnson J L, Padgett M L. PCNN models and applications[J]. IEEE Trans. on Neural Networks.1999,10 (3):480-498.
[10]马义德.生物信息的数字图像处理技术研究[博士论文].兰州：兰州大学.2001.
[11]张军英,卢志军,石林,董继扬,石美红.基于脉冲耦合神经网络的椒盐噪声图像滤波[J].中国科学E辑.2004,34(8)：882-894.
[12]刘勃,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,(08)：1869-1873.
[13]卢桂馥,王勇,窦易文.一种新的基于PCNN的图像脉冲噪声滤波算法[J].计算机技术与发展.2007,17(12)：83-85.
[14]马义德,张红娟PCNN与灰度形态学相结合的图像去噪方法[J].北京邮电大学学报.2008,31(02)：108-112.
[15]王巍,黎明,刘高航.基于脉冲耦合神经网络的彩色图像滤波新方法[J].计算机工程与设计.2007,28(14)：3413-3415.
[16]陈锐,马义德,郑晓,袁树林.基于PCNN和视觉特性的图像对比度增强[J］.微计算机信息.2010,26(4-1)：108-111.
[17]徐光柱,刘鸣,任东,马义德,刘晓丽.基于脉冲耦合神经网络的多区域图像分割[J].山东大学学报(理学版).2010,45(07)：86-93.
[18]张北斗,马义德,林冬梅,张立文.基于脉冲耦合神经网络的超声图像病理区域检测算法[J].兰州大学学报(自然科学版).2008,44(01)：61-64.
[19]绽琨,张红娟,马义德,刘丽,田乐.交叉皮层模型及其在图像处理中的应用[J].北京邮电大学学报.2009,32(04)：40-45.
[20]李志强,程飞燕,安黎哲,马义德,高清祥.基于多维PCNN的彩色枸杞细胞显微图像 的边缘检测[J].兰州大学学报(自然科学版).2009,45(05)130-134.
[21]刘映杰,马若飞,朱望飞,者吴,绽琨,马义德.基于多阈值PCNN的运动目标检测算法[J].计算机应用.2009,29(03)：739-741.
[22]朱望飞,马义德,邱秀清.基于PCNN的高斯混合模型运动检测改进方法[J].兰州大学学报(自然科学版).2009,45(02)：129-131.
[23]王新PCNN混沌特性与硬件实现研究[硕士论文].兰州：兰州大学.2010.
[24]赵釜.基于Cyclone Ⅱ系列FPGA的图像实时采集与预处理系统研究[硕士论文].重庆：重庆大学.2009.
[25]张志刚FPGA与SOPC设计教程——DE2实践[M].西安：西安电子科技大学出版社.2007.4：9-30.
[26]仙云森.基于FPGA的图像处理算法研究及硬件设计[硕士论文].大连：大连理工大学.2008.
[27]Altera Corporation. Quartus Ⅱ Handbook [EB/OL]. http://www.altera.com.
[28]Terasic Technologies Inc. DE2-70 User manual [EB/OL]. http://www.terasic.com.cn.
[29]Altera Corporation. Altera DE2-70 Development and Education Board User Manual [EB/OL]. http://www.altera.com.
[30]Altera Corporation. De2-70 Handbook [EB/OL]. http://www.altera.com.
[31]史飞.脉冲耦合神经网络在图像处理中的应用研究[硕士论文].兰州：兰州大学.2003.
[32]李洋波,赵不贿.基于FPGA的图像中值滤波器的硬件实现[J].现代电子技术.2008,(22)：99-101.
[33]马义德,李廉,绽锟,王兆斌.脉冲耦合神经网络与数字图像处理[M].北京：科学出版社.2008.7.
[34]付昱强.基于FPGA的图像处理算法的研究与硬件设计[硕十论文].南昌：南昌大学.2006.
[35]刘健鹏,陈卫东,钱钧.基于FPGA的实时中值滤波器设计[J].应用光学.2007,28(06)：712-715.
[36]郑鹤,王鲁平,李飙.一种快速的二维中值滤波算法及其硬件实现[J].信息与电子工程.2005,3(04)：245-248.
[37]李轶博,李小兵,周娴.基于FPGA的快速中值滤波器设计与实现[J].液晶与显示.2010,25(02)：292-296.
[38]李飞飞,刘伟宁,王艳华.改进的中值滤波算法及其FPGA快速实现[J].计算机工程.2009,35(14)：175-177.
[39]陈加成,徐熙平,吴琼.基于FPGA的中值滤波算法研究与硬件设计[J].长春理[大学学报(自然科学版).2008,31(01)8-10.
[40]黄艳军.基于FPGA的数字图像预处理算法研究[硕十论文].南京：南京理工大学.2009.
[41]徐大鹏,李从善.基于FPGA的数字图像中值滤波器设计[J].电子器件.2006,29(04)1114-1117.
[42]胡越黎,计慧杰,吴频,宣祥光.图像的中值滤波算法及其FPGA实现[J].计算机测量与控制.2008,16(11)：1672-1675.