1 简介
1950s数字图像处理产生，从此学术界对它的研究连绵不绝。十多年之后，数字图像的处理被独立出来成了一门课程。图像分割技术是处理图像的一种基础技术，在图像处理领域中占了重要地位，且在图像工程、模式识别、计算机视觉等方面起着重要作用。本文正是以图像分割为研究课题，以如何能更好地图像进行省时而有效的分割为研究目的，提出了基于最大熵的脉冲耦合神经网络（PulseCoupled Neural Networks，简称PCNN）的方法。众所周知，有效的分割方法有利于理解图像的本质和后续的研究工作等，因此本文分割方法的研究是具有一定意义的。 传统的处理图像的分割问题的方法有：基于灰度级算术均值法、基于熵与直方图法、基于最大类间方差法、基于边缘的检测分割法，基于阈值的分割法，基于边界提取的分割法，基于区域的分割法，基于特定理论的分割法。同时，当然少不了产生的一些相关模型，如健壮尺度区域模型，二维经典C-V模型，尺度区域拟合模型，三维C-V模型，健壮统计三维C-V模型，几何主动轮廓模型等。本文将出基于脉冲耦合神经网络的方法。此方法对上述分割数字图像时的不足具有一定的弥补作用。用PCNN方法分割数字图像的模型中，以神经元为单元，构成二维的一层的神经元列。PCNN模型中神经元的数目一致于像素数目，每个神经元一一对应于每个像素。根据PCNN的脉冲传播特性而引起的同步脉冲现象来实现图像分割。
2 部分代码
clear all
RGB = imread('park.bmp');
figure(1);
imshow(RGB);
imwrite(RGB,'分割结果\park.jpg','jpg');
HSV = rgb2hsv(RGB); % Transform from RGB to HSV
H = HSV(:,:,1);
S = HSV(:,:,2);
V = HSV(:,:,3);
H = H+0.22;
INDEX = find(H>1);
H(INDEX) = H(INDEX) - 1;
figure(24);
imshow(H);
imwrite(H,'分割结果\park_H_revolve.jpg','jpg');
% Color quantization
QH = 16;
QS = 4;
QV = 4;
scopeH = 1 / QH;
scopeS = 1 / QS;
scopeV = 1 / QV;
siz = size(H);
M = siz(1) * siz(2);
temp = zeros(siz);
HHH = temp;
SSS = temp;
VVV = temp;
% Quantize H
for i = 1:QH
k = find((H < i*scopeH) & (H >= (i-1)*scopeH));
HHH(k) = i;
end
% Quantize S
for i = 1:QS
k = find((S < i*scopeS) & (S >= (i-1)*scopeS));
SSS(k) = i;
end
% Quantize V
for i = 1:QV
k = find((V < i*scopeV) & (V >= (i-1)*scopeV));
VVV(k) = i
end
% Color label
QI = temp; % label matrix, used to statistic Ck
for i = 1:siz(1)
for j = 1:siz(2)
QI(i,j) = (HHH(i,j) -1)*QS*QV + (SSS(i,j) - 1)*QV + VVV(i,j);
end
end
eQI = uint8(QI);
figure(35);
imshow(eQI);
imwrite(eQI,'分割结果\park_quantitation.jpg','jpg');
%imwrite(eQI,['Lajiao_VQofHSV','.bmp'],'bmp');
%figure(112);
%EDG2 = edge(eQI,'canny');
%imshow(EDG2);
%imwrite(EDG2,['Lajiao_VQofHSV_edg','.bmp'],'bmp');
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
% Double PCNN
%%%link parameter%%
[row, col] = size(QI);
Va = max(max(QI));
Vb = min(min(QI));
F = QI;
vl = 1;
vt = 500;
l_deta = 1;
l_t = 0.5;
link_a = l_deta*1 / l_t;
beta = 0.012;
t_deta = 1;
t_t = 25;
threshold = t_deta*1 / t_t;
%step = 20;%optimize
step = 20;
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
TEMP=zeros(row,col);
%%%%%%%%%%% to create W %%%%%%%%%%%%%%
%Ws=[0 1 0;1 1 1;0 1 0];
%%%%%%%%%%%%%%%%%%%%%
radius=9;
halfR = round(radius/2);
deta = 2;
for i = 1:radius
for j = 1:radius
if i==halfR & j==halfR
K_r(halfR, halfR)=1;
K_r2(halfR, halfR)=1;
else
K_r(i,j) = 1/sqrt((i-halfR)^2 + (j-halfR)^2);
end
end
end
Ws = K_r;
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
%%%%%%% start program %%%%%%%%%%%%%
%%%% factor loop %%%%%
Y_threshold = TEMP;
Y_time = TEMP;
Y1 = TEMP;
Y2 = TEMP;
Y = TEMP;
Ya = TEMP;
Yb = TEMP;
Edge_image=TEMP;
L = TEMP;
U = TEMP;
T1 = TEMP + Va;
T2 = TEMP + Vb;
j = 1;
%%% determine when exit loop
accuYTrue = 1;
iterTrue = 1;
accY = TEMP; % Accumulate total neuros of firing
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
while iterTrue
j
L = link24(Y,L,Ws,link_a,vl);
invar_fig=1;change_mark=0;k=0;
%%%%%%%%%%%%%%%
m=1;%% fast linking %%
while (invar_fig==1)
m
mid_Y=Y;
U=internal24(F,L,Y,beta);
Y1 = pulse1(U, T1);
%Y2 = pulse2(U, T2);
Y = Y1; % + Y2;
%Y=pulse_p(U,T,L);
if (mid_Y==Y)
invar_fig=0;
elseif m>30 & change_mark==0
mid_Y1=mid_Y;
mid_Y2=Y;
change_mark=1;
elseif change_mark==1 & k<1
k=1;
elseif k==1
%if mid_Y1==mid_Y & mid_Y2==Y
invar_fig=0;
%else
% change_mark=0;k=0;
% end
end
L=link24(Y,L,Ws,link_a,vl);
m=m+1;
end
%%%%save threshold for fired pixels (sigle-pass)%%%%%%
%%%%%%%statistc numbers of nurons in plusing areas %%%%%%
index1 = find(Y1 ~= 0);%find index of element of noequal zero(index of pulsing neurons)
q = size(index1, 1);
if q ~= 0 %%% statistic pulsing neurons %%%
for yy = 1:row
for zz = 1:col
if(Y1(yy, zz) == 1)
Y_threshold(yy, zz) = round(T1(yy, zz));
Y_time(yy, zz) = j;
end
end
end %%% statistic end %%%
end
Ya = Ya + Y1;
T1 = threshold1(T1, Ya, Va, step); %decrement threshold(linear decay)%%%%%%%%%%
j=j+1;
%%%%%%%%%%%%%%%%%%%%%
accY = accY + Y;
index = find(accY == 0);
size(index);
if ans(1) == 0
iterTrue = 0;
end % Exit loop
end
figure(59);
Y_pcnn = uint8(Y_threshold)
imshow(Y_pcnn);
imwrite(Y_pcnn,'分割结果\park_pcnn_segm2.bmp','bmp');
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
3 仿真结果
4 参考文献
[1]顾晓东, 余道衡. PCNN的原理及其应用[J]. 电路与系统学报, 2001(03):46-51.
博主简介：擅长智能优化算法、神经网络预测、信号处理、元胞自动机、图像处理、路径规划、无人机等多种领域的Matlab仿真，相关matlab代码问题可私信交流。
部分理论引用网络文献，若有侵权联系博主删除。
