基于生物侧抑制机制的神经网络模型研究
|
摘要
侧抑制机制存在于生物神经系统中,是神经系统信息处理的基本原则之一,能够运用在图像处理的各个方面,使得处理结果图更符合“人眼”的视觉效果。基于生物生理特性,研究者们提出很多神经网络模型。本文将从侧抑制机制出发,探讨研究基于视觉系统的侧抑制构成的神经网络模型。
首先,研究基于生物视觉系统提出的脉冲耦合神经网络(Pulse Coupled Neural Network,PCNN),PCNN运用于图像处理是一种单层二维的局部连接网络,神经元与像素点一一对应。本文对PCNN进行参数讨论,并将其运用于图像分割、图像增强、边缘检测;与传统方法比较优劣,基于生物视觉系统的模型用于图像处理具有运行速度较快、结果比较自然、灰度表现力好、且能够很好地提取图像中的信息等特点。
其次,对生物侧抑制模型进行分类并讨论,针对侧抑制模型运用于图像处理方面的缺陷提出两种新的方法:一种是加入可能性度量因子;一种是与自适应滤波相结合;进行实验分析论证方法的可行性。接着针对PCNN在图像处理方面的缺陷,给出一种新的模型,即侧抑制机制与PCNN相结合的模型进行仿真实验,所得的结果比单纯PCNN仿真结果轮廓线条更加明确,图像具有更好的连通性。
最后,给出一种基于点火(Integrate-and-Fire,IF)的侧抑制机制神经网络群用于位置跟踪的模型,在此基础上,提出一种新的迭代训练算法PITS(Progressive interactive training scheme,PITS)算法进行参数学习,利用信息中心(Information Center,IC)储存每次训练结果,在保证收敛的情况下,比较跟踪结果的误差函数给出权值调整公式进行自学习。运用H-H简化的模型在模拟侧抑制机制的同时实现位置跟踪,新的学习算法提高了跟踪精度和速度。
Lateral inhibition is one of the basic principles in neural information processing systems which exist in biological neural systems. It can be used in image processing to make the result more suitable to human visual system. Many neural network models have been proposed based on the biological physiological characteristics. This paper will make a research on the neural network model based on lateral inhibition of the visual system .
First, the pulse coupled neural network (PCNN), which is based on the biological system, has been researched. In image processing, PCNN is used as a single 2-D local connection network, where nerve cell and pixel are one to one correspondence. This paper discusses the parameters of PCNN and its application to image segmentation, image enhancement and edge detection. Advantages and disadvantages are compared with traditional methods to study the visual system model based on biological characteristics. The model based on biological vision systems in image processing runs faster, performs more natural results, and extracts information well from images with good grayscale expression.
Second, classification and discussion of biological lateral inhibition model have been conducted. Two new methods have been proposed according to the weakness of the lateral inhibition model used in image processing: one is added a possibility measure factor; the other is combined with the adaptive filter. Experiments demonstrate the feasibility of the two methods. Based on the weakness of PCNN in image processing, a new model-the combination of lateral inhibition mechanism and the PCNN model is proposed. Simulation has been made to compare with the pure PCNN model. The results have more clearly contours and better connectivity.
Finally, a location tracking model based on Integrate-and-Fire (IF) mechanism using neural network ensemble of lateral inhibition has been proposed. A new iterative training algorithm PITS (Progressive iterative training scheme) is used for parameters learning. Using information center (IC) to store the results of each training session. To ensure convergence , the tracking weight adjustment formula is given by comparing the results of the error function. The IF model which is simplified by H-H model is used. The lateral inhibition is simulated meanwhile the purpose of position tracking is achieved, and the tracking accuracy and speed have been increased because of the new learning algorithm.
首先,研究基于生物视觉系统提出的脉冲耦合神经网络(Pulse Coupled Neural Network,PCNN),PCNN运用于图像处理是一种单层二维的局部连接网络,神经元与像素点一一对应。本文对PCNN进行参数讨论,并将其运用于图像分割、图像增强、边缘检测;与传统方法比较优劣,基于生物视觉系统的模型用于图像处理具有运行速度较快、结果比较自然、灰度表现力好、且能够很好地提取图像中的信息等特点。
其次,对生物侧抑制模型进行分类并讨论,针对侧抑制模型运用于图像处理方面的缺陷提出两种新的方法:一种是加入可能性度量因子;一种是与自适应滤波相结合;进行实验分析论证方法的可行性。接着针对PCNN在图像处理方面的缺陷,给出一种新的模型,即侧抑制机制与PCNN相结合的模型进行仿真实验,所得的结果比单纯PCNN仿真结果轮廓线条更加明确,图像具有更好的连通性。
最后,给出一种基于点火(Integrate-and-Fire,IF)的侧抑制机制神经网络群用于位置跟踪的模型,在此基础上,提出一种新的迭代训练算法PITS(Progressive interactive training scheme,PITS)算法进行参数学习,利用信息中心(Information Center,IC)储存每次训练结果,在保证收敛的情况下,比较跟踪结果的误差函数给出权值调整公式进行自学习。运用H-H简化的模型在模拟侧抑制机制的同时实现位置跟踪,新的学习算法提高了跟踪精度和速度。
Lateral inhibition is one of the basic principles in neural information processing systems which exist in biological neural systems. It can be used in image processing to make the result more suitable to human visual system. Many neural network models have been proposed based on the biological physiological characteristics. This paper will make a research on the neural network model based on lateral inhibition of the visual system .
First, the pulse coupled neural network (PCNN), which is based on the biological system, has been researched. In image processing, PCNN is used as a single 2-D local connection network, where nerve cell and pixel are one to one correspondence. This paper discusses the parameters of PCNN and its application to image segmentation, image enhancement and edge detection. Advantages and disadvantages are compared with traditional methods to study the visual system model based on biological characteristics. The model based on biological vision systems in image processing runs faster, performs more natural results, and extracts information well from images with good grayscale expression.
Second, classification and discussion of biological lateral inhibition model have been conducted. Two new methods have been proposed according to the weakness of the lateral inhibition model used in image processing: one is added a possibility measure factor; the other is combined with the adaptive filter. Experiments demonstrate the feasibility of the two methods. Based on the weakness of PCNN in image processing, a new model-the combination of lateral inhibition mechanism and the PCNN model is proposed. Simulation has been made to compare with the pure PCNN model. The results have more clearly contours and better connectivity.
Finally, a location tracking model based on Integrate-and-Fire (IF) mechanism using neural network ensemble of lateral inhibition has been proposed. A new iterative training algorithm PITS (Progressive iterative training scheme) is used for parameters learning. Using information center (IC) to store the results of each training session. To ensure convergence , the tracking weight adjustment formula is given by comparing the results of the error function. The IF model which is simplified by H-H model is used. The lateral inhibition is simulated meanwhile the purpose of position tracking is achieved, and the tracking accuracy and speed have been increased because of the new learning algorithm.
引文
[1]陈卉.用于图像增强的侧抑制网络模型的仿真比较[J].系统仿真学报.2003.15(1):100-103
[2]周常河,刘立人,王之江.Hartline侧抑制神经网络模型的光学实验[J].光学学报.1992.12(5):451-456.
[3]GRAHAM.N,RATLIFF.F,HARTLINE.HK.Facilitation of inhibition in compound lateral eye of limulus[J]. Proceedings of the national academy of sciences of the united states of America.1973:70(3): 894-898.
[4]杜红伟.生物侧抑制机制及应用研究[D].南京:南京航空航天大学,2008.
[5]Bowes,.D, Adams.R,Canamero.L, et al. The role of lateral inhibition in the sensory processing in a simulated spiking neural controller for a robot[C]. 2009 IEEE Symposium on artifical life.2009: 179-183.
[6]王红梅,李言俊,张科.生物视觉仿生在计算机视觉中的应用研究[J].计算机应用研究.2009.26(3):1157-1159.
[7]李言俊.视觉仿生成像制导技术及应用[M].北京:国防工业出版社,2006.
[8]王从庆,杜红伟.一种基于皮层柱侧抑制机制的神经网络群[J].信息与控制.2008. 37(4): 413-417.
[9]杜红伟,王从庆.基于演算侧抑制模型的运动目标检测方法[J].湘潭大学自然科学学报.2009. 31(3):143-147.
[10]Zhang. D, Nishimura. TH.Pulse coupled neural network based anisotropic diffusion method for 1/f noise reduction[J]. Mathematical and computer modelling.2010.52(11-12):2085-2096.
[11]Chen.QH, Zhou. L.A Pulse-Coupled Neural Network for Color Image Edge Detection[C].Proceedings of the 8th wseas international conference on applied computer and applied computational science. 2009:441-445.
[12]陈锐,马义德,郑晓.基于PCNN和视觉特性的图像对比度增强[J].微计算信息.2010.26(4-1): 108-112.
[13]Charles M.Gray,Andreas K.Engel.Oscillatory responses in cat visual cortex exhibit inter-columnar synchronization which reflects global stimulus properties[J].Nature 338.1989.3(23).334-337.
[14]顾晓东,余道衡.PCNN的原理及应用[J].电路与系统学报.2010.6(3):45-50.
[15] Kuntimad G,Ranganath H S.Perfect image segmentation using pulse-coupled neural network[J]. IEEE Transaction on Neural Networks,1999.10(3):591-598.
[16]马义德,张恩溯.基于非抽样contourlet和PCNN的图像凹点检测方法研究[J].系统仿真学报.2008.20(10):2579-2583.
[17]Zhaobin Wang,Yide Ma,Feiyan Cheng.Review of pulse-coupled neural networks[J].Image and Vision Computing.2010.28(1):5-13.
[18]马义德,脉冲耦合神经网络原理及应用[M].北京:科学出版社,2006.
[19]邓灵博.基于PCNN和最大相关准则的图像分割算法[J].弹箭与制导学报.2008.28(3):237-239
[20]Eckhorn R,Reitboeck HJ,Arndt M,et al.Feature linking via synchronization among distributed assemblies:Simulation of result from cat visual cortex[J].Neutral Computation,1990,2(3):293-307.
[21]Ranganath HJ,Kuntimad G,Johnson JL.Pulse coupled neural networks for image processing[C]. PCNN International Workshop,Huntsville:IEEE.1995:37-43
[22]Ben Ayed.I,Li.S,Ross.I.Level Set Image Segmentation with a Statistical Overlap Constraint [J].Information Processing in Medical Imaging,Proceedings.2009.5636:589-601.
[23]肖飞等.图像分割方法综述[J].可编程控制器与工厂自动化.2009.11:77-79.
[24]顾晓东,余道衡,郭仕德.关于PCNN应用于图像处理的研究[J].电讯技术.2003.43(3):21-24.
[25]LI Gang , ZANG Rui li, LIN Ling . Research on vehicle license plate location based on neural networks[ C] . Washington DC: Proceedings of the First International Conference on Innovative Computing , Information and Control. 2006.3:174-177.
[26]Du.S,R.K.Ward. Adaptive Region-Based Image Enhancement Method for Robust Face Recogniti- On Under Variable Illumination Conditions[J].IEEE Transaction on Circuits and Systems for Video Technology.2010.20:1165-1175.
[27]马振兴.基于脉冲耦合神经网络的医学图像增强处理[D].北京:北京邮电大学,2008.
[28]石美红,张军英,李永刚.等.一种新的低对比度图像增强方法[J].计算机应用研究.2005.1:235- 238.
[29]孙存杰,张虹.CT图像增强方法的研究[J].CT理论与应用研究.2005.24(4):17-20.
[30]Krantz J H.A computational model of the retina Presented at the 30th annual meeting of the Society for Computers in Psychology[J].New Orleans,2000,LA.
[31]赵峙江,张田文,张志宏.一种基于视觉模型与连通域统计的阈值分割新算法[J].电子学报.2005.33(5):793-797.
[32]王立,张科,李言俊.视网膜皮层图像信号处理的传输机制与仿真[J].光学技术.2003.29(4):420-422.
[33]刘怀贤,姚晓东,常青.侧抑制网络在图像增强中的应用[J].华东理工大学学报(自然科学版).2007.33(1):120-123.
[34]谢晓芳,毛晓波,陈铁军.改进的侧抑制网络图像增强算法的研究[J].计算机工程与应用.2010.46(3):148-150.
[35]Aburas H M,Cetiner B G,Sari M.Dengue confirmed-cases prediction:a neural network model [J].Expert Systems with Applications.2010.37(6):4250-4260.
[36]Dehuri S,Cho S B.Evolutionarily optimized features in functional link neural network for classification[J].Expert Systems w ith Applications.2010.37(6):4379-4391.
[37]Elvin A J,Laing C R,Mclachlan R I.Exploiting the Hamiltonian structure of a neural field model[J].Physica D-Nonlinear Phenomena.2010.239(9):537-546.
[38]L.F.Abbott.Lapicque’s introduction of the integrate-and-fire model neuron(1907)[J].Brain Research Bulletion.1999.50(5/6):303-304.
[39]Akhand M A H,Islam M M,Murase K.Progressive interactive training:a sequential neural network ensemble learning method[J].Neurocomuputing.2009.73(1/2/3):260-273.
[40]T.H.Bullock,et al.Introduction to Nervous System[M].San Francisco:Freeman&Co..1977.
[41]王岳环,陈妍,程胜莲,等.基于模式侧抑制复杂背景下的小目标检测[J].红外与激光工程.2005.34(6):703-708.
[42]Satish Kumar.Neural Networks[M].北京:清华大学出版社.2006.
[43]Brown G,Wyatt J,Harris R,et al.Diversity creation methods:a survey and categorization[J].Inform- ation Fusion,2005.6(1):99-111.
[44]梁爽,王从庆.一种基于IF模型侧抑制神经网络群的PITS学习算法[J].东南大学学报(自然科学版).2010.41(1):178-182.
[45]赵璇,何波,吉德志,等.基于粒子滤波的机器人定位及动态目标跟踪[J].系统仿真学报.2008.20(3):6490-6497.
[2]周常河,刘立人,王之江.Hartline侧抑制神经网络模型的光学实验[J].光学学报.1992.12(5):451-456.
[3]GRAHAM.N,RATLIFF.F,HARTLINE.HK.Facilitation of inhibition in compound lateral eye of limulus[J]. Proceedings of the national academy of sciences of the united states of America.1973:70(3): 894-898.
[4]杜红伟.生物侧抑制机制及应用研究[D].南京:南京航空航天大学,2008.
[5]Bowes,.D, Adams.R,Canamero.L, et al. The role of lateral inhibition in the sensory processing in a simulated spiking neural controller for a robot[C]. 2009 IEEE Symposium on artifical life.2009: 179-183.
[6]王红梅,李言俊,张科.生物视觉仿生在计算机视觉中的应用研究[J].计算机应用研究.2009.26(3):1157-1159.
[7]李言俊.视觉仿生成像制导技术及应用[M].北京:国防工业出版社,2006.
[8]王从庆,杜红伟.一种基于皮层柱侧抑制机制的神经网络群[J].信息与控制.2008. 37(4): 413-417.
[9]杜红伟,王从庆.基于演算侧抑制模型的运动目标检测方法[J].湘潭大学自然科学学报.2009. 31(3):143-147.
[10]Zhang. D, Nishimura. TH.Pulse coupled neural network based anisotropic diffusion method for 1/f noise reduction[J]. Mathematical and computer modelling.2010.52(11-12):2085-2096.
[11]Chen.QH, Zhou. L.A Pulse-Coupled Neural Network for Color Image Edge Detection[C].Proceedings of the 8th wseas international conference on applied computer and applied computational science. 2009:441-445.
[12]陈锐,马义德,郑晓.基于PCNN和视觉特性的图像对比度增强[J].微计算信息.2010.26(4-1): 108-112.
[13]Charles M.Gray,Andreas K.Engel.Oscillatory responses in cat visual cortex exhibit inter-columnar synchronization which reflects global stimulus properties[J].Nature 338.1989.3(23).334-337.
[14]顾晓东,余道衡.PCNN的原理及应用[J].电路与系统学报.2010.6(3):45-50.
[15] Kuntimad G,Ranganath H S.Perfect image segmentation using pulse-coupled neural network[J]. IEEE Transaction on Neural Networks,1999.10(3):591-598.
[16]马义德,张恩溯.基于非抽样contourlet和PCNN的图像凹点检测方法研究[J].系统仿真学报.2008.20(10):2579-2583.
[17]Zhaobin Wang,Yide Ma,Feiyan Cheng.Review of pulse-coupled neural networks[J].Image and Vision Computing.2010.28(1):5-13.
[18]马义德,脉冲耦合神经网络原理及应用[M].北京:科学出版社,2006.
[19]邓灵博.基于PCNN和最大相关准则的图像分割算法[J].弹箭与制导学报.2008.28(3):237-239
[20]Eckhorn R,Reitboeck HJ,Arndt M,et al.Feature linking via synchronization among distributed assemblies:Simulation of result from cat visual cortex[J].Neutral Computation,1990,2(3):293-307.
[21]Ranganath HJ,Kuntimad G,Johnson JL.Pulse coupled neural networks for image processing[C]. PCNN International Workshop,Huntsville:IEEE.1995:37-43
[22]Ben Ayed.I,Li.S,Ross.I.Level Set Image Segmentation with a Statistical Overlap Constraint [J].Information Processing in Medical Imaging,Proceedings.2009.5636:589-601.
[23]肖飞等.图像分割方法综述[J].可编程控制器与工厂自动化.2009.11:77-79.
[24]顾晓东,余道衡,郭仕德.关于PCNN应用于图像处理的研究[J].电讯技术.2003.43(3):21-24.
[25]LI Gang , ZANG Rui li, LIN Ling . Research on vehicle license plate location based on neural networks[ C] . Washington DC: Proceedings of the First International Conference on Innovative Computing , Information and Control. 2006.3:174-177.
[26]Du.S,R.K.Ward. Adaptive Region-Based Image Enhancement Method for Robust Face Recogniti- On Under Variable Illumination Conditions[J].IEEE Transaction on Circuits and Systems for Video Technology.2010.20:1165-1175.
[27]马振兴.基于脉冲耦合神经网络的医学图像增强处理[D].北京:北京邮电大学,2008.
[28]石美红,张军英,李永刚.等.一种新的低对比度图像增强方法[J].计算机应用研究.2005.1:235- 238.
[29]孙存杰,张虹.CT图像增强方法的研究[J].CT理论与应用研究.2005.24(4):17-20.
[30]Krantz J H.A computational model of the retina Presented at the 30th annual meeting of the Society for Computers in Psychology[J].New Orleans,2000,LA.
[31]赵峙江,张田文,张志宏.一种基于视觉模型与连通域统计的阈值分割新算法[J].电子学报.2005.33(5):793-797.
[32]王立,张科,李言俊.视网膜皮层图像信号处理的传输机制与仿真[J].光学技术.2003.29(4):420-422.
[33]刘怀贤,姚晓东,常青.侧抑制网络在图像增强中的应用[J].华东理工大学学报(自然科学版).2007.33(1):120-123.
[34]谢晓芳,毛晓波,陈铁军.改进的侧抑制网络图像增强算法的研究[J].计算机工程与应用.2010.46(3):148-150.
[35]Aburas H M,Cetiner B G,Sari M.Dengue confirmed-cases prediction:a neural network model [J].Expert Systems with Applications.2010.37(6):4250-4260.
[36]Dehuri S,Cho S B.Evolutionarily optimized features in functional link neural network for classification[J].Expert Systems w ith Applications.2010.37(6):4379-4391.
[37]Elvin A J,Laing C R,Mclachlan R I.Exploiting the Hamiltonian structure of a neural field model[J].Physica D-Nonlinear Phenomena.2010.239(9):537-546.
[38]L.F.Abbott.Lapicque’s introduction of the integrate-and-fire model neuron(1907)[J].Brain Research Bulletion.1999.50(5/6):303-304.
[39]Akhand M A H,Islam M M,Murase K.Progressive interactive training:a sequential neural network ensemble learning method[J].Neurocomuputing.2009.73(1/2/3):260-273.
[40]T.H.Bullock,et al.Introduction to Nervous System[M].San Francisco:Freeman&Co..1977.
[41]王岳环,陈妍,程胜莲,等.基于模式侧抑制复杂背景下的小目标检测[J].红外与激光工程.2005.34(6):703-708.
[42]Satish Kumar.Neural Networks[M].北京:清华大学出版社.2006.
[43]Brown G,Wyatt J,Harris R,et al.Diversity creation methods:a survey and categorization[J].Inform- ation Fusion,2005.6(1):99-111.
[44]梁爽,王从庆.一种基于IF模型侧抑制神经网络群的PITS学习算法[J].东南大学学报(自然科学版).2010.41(1):178-182.
[45]赵璇,何波,吉德志,等.基于粒子滤波的机器人定位及动态目标跟踪[J].系统仿真学报.2008.20(3):6490-6497.