Research On Pulse Coupled Neural Networks With Self-adapting Parameter Setting And Its Application In Image Processing

The Pulse Coupled Neural Networks model(PCNN)was presented at the end of the 1990 s by Johnson et al.,based on the Eckhorn's mammalian visual cortex model.The model has a nonlinear modulation coupling characteristic,nonlinear dynamic threshold attenuation characteristic,dynamic pulse burst characteristic and synchronous pulse burst characteristic.It is a new type of artificial neural network model which is different from traditional artificial neural network.It is considered as the third generation artificial Neural Networks.PCNN has good characteristics of rotation invariance,scale invariance,distortion invariance,and signal intensity invariance.PCNN model is a good simulation of biological visual nervous system,more in line with human visual characteristics,in image processing,computer vision has a unique advantage.In this paper,the Pulse Coupled Neural Networks model is systematically studied deeply.Based on the analysis of the firing mechanism and characteristics of PCNN,Intersecting Cortical Model(ICM)and Spiking Cortical Model(SCM).An Improved Pulse Coupled Neural Networks model(IPCNN)and a Heterogeneous Pulse Coupled Neural Network model(HIPCNN)are proposed.The two methods are applied to image processing,such as image segmentation,edge detection and so on.The Berkeley Segmentation Dataset are used to test the algorithm,and the experimental result prove the validity and efficiency.In this paper,the main research works as follows:1.The development process,research status and problems of PCNN are introduced.The basic structure of PCNN and the improved model ICM and SCM are introduced.The firing mechanism and characteristics of PCNN model are analyzed emphatically.And briefly analyzes the function of each parameter in PCNN model.In the application of the red blood cell count,a new method based on PCNN and image quality is proposed,which extracts the contours of RBC from various microscopic images of disparate blood smears,it is efficient and accurate,it is a viable practical method.2.An IPCNN model is proposed,which reduces the number of parameters and maintains the feeding input and linking input mode of PCNN model.The dynamic characteristics and the firing mechanism of IPCNN model are studied in depth.The relationship between the static characteristics of the input image and the dynamic characteristics of the neuron is constructed,so that all four parameters of the IPCNN model are set adaptively,which avoids the manual setting method and improves the parameter adaptability,and the method does not require training and preexperimentation;3.Based on the IPCNN model,a HIPCNN model is proposed.The fixed linking coefficient of the model is modulated according to the static characteristics of the image,and the variable linking coefficient is constructed.The innovative heterogeneous network is introduced into the pulse coupled neural network;4.By applying IPCNN model and HIPCNN model into the image processing area of image segmentation,the model parameters are adaptively set according to the static characteristics of the input image,and the number of iterations is very small,it is efficient and accurate.In the image processing area of edge detection,a two stage model based on IPCNN / HIPCNN and PCNN is proposed,which can easily and efficiently obtain the edge information of the image.