车牌角度校正.doc

function picbw_xz,angle_xz=radon(temparea_gray)%temparea_gray=rgb2gray(temparea);height,width=size(temparea_gray);level=ostugetT(temparea_gray,height,width);picbw=im2bw(temparea_gray,level/255.0);%figure(1);imshow(picbw);sum_xz=zeros(21,height);total_xz=zeros(1,21);max_total=0;for angle=-5:5picbw_temp = imrotate(picbw,angle,crop);t=angle+11;%for i=1:height sum_xz(t,i)=0; for j=1:width sum_xz(t,i)=sum_xz(t,i)+picbw_temp(i,j); endendtotal_xz(t)=0;for i=1:height-1 total_xz(t)=total_xz(t)+abs( sum_xz(t,i)-sum_xz(t,i+1) ); endif total_xz(t)max_total angle_xz=angle; picbw_xz=picbw_temp; max_total=total_xz;endend 最小二乘我没用过；倾斜校正用得比较多的是hough变换和randon变换；hough变换必须得在车牌二值化的时候出现边框，要是粗定位没有边框或本来车牌边框不明显就没辙了。所以貌似radon变换比较靠谱；也叫旋转投影法，鲁棒性也比较好。我现在用的是改进的radon变换，以检测投影值的一阶导数累加值为基础。你可以先看看相关论文下面是我的radon变换子程序，你可以参考一下function picbw_xz,angle_xz=radon(temparea_gray)%temparea_gray=rgb2gray(temparea);height,width=size(temparea_gray);level=ostugetT(temparea_gray,height,width);picbw=im2bw(temparea_gray,level/255.0);%figure(1);imshow(picbw);sum_xz=zeros(21,height);total_xz=zeros(1,21);max_total=0;for angle=-5:5picbw_temp = imrotate(picbw,angle,crop);t=angle+11;%for i=1:height sum_xz(t,i)=0; for j=1:width sum_xz(t,i)=sum_xz(t,i)+picbw_temp(i,j); endendtotal_xz(t)=0;for i=1:height-1 total_xz(t)=total_xz(t)+abs( sum_xz(t,i)-sum_xz(t,i+1) ); endif total_xz(t)max_total angle_xz=angle; picbw_xz=picbw_temp; max_total=total_xz;endend 字符识别的话用BP神经网络好了，简单。下面这段是用来训练神经网络的，BP网络的基础知识是必须的啊忍忍多看两遍就理解了。文件夹下面有1199张阿拉伯数字的样本图片，循环送到MATLAB中保存3216200的特征值，并与输出教师信号t对应,存在num_PT里头。然后送到matlab的神经网络工具箱里面去训练。（相关资料百度文库里面也有）最后保存网络 % 引自神经网络模型及其MATLAB仿真程序设计一书P9498% 生成输入向量和目标向量clear all;LOADING.for kk = 0:199 m=strcat(use_num,int2str(kk),.jpg);% x=imread(m,jpg); %依次输入训练字符。 bw1=im2bw(x,0.5); %用0.5阈值进行二值化 for m=0:15 p(m*32+1:(m+1)*32,kk+1)=bw1(1:32,m+1); end switch kk case0,1,2,3,4,5,6,7,8,9,100,101,102,103,104,105,106,107,108,109 t(:,kk+1)=0 0 0 0 0 0 0; % t(kk+1)记录了真实的数值 case10,11,12,13,14,15,16,17,18,19,110,111,112,113,114,115,116,117,118,119 t(:,kk+1)=0 0 0 0 0 0 1; case20,21,22,23,24,25,26,27,28,29,120,121,122,123,124,125,126,127,128,129 t(:,kk+1)=0 0 0 0 0 1 0; case30,31,32,33,34,35,36,37,38,39,130,131,132,133,134,135,136,137,138,139 t(:,kk+1)=0 0 0 0 0 1 1; case40,41,42,43,44,45,46,47,48,49,140,141,142,143,144,145,146,147,148,149 t(:,kk+1)=0 0 0 0 1 0 0; case50,51,52,53,54,55,56,57,58,59,150,151,152,153,154,155,156,157,158,159 t(:,kk+1)=0 0 0 0 1 0 1; case60,61,62,63,64,65,66,67,68,69,160,161,162,163,164,165,166,167,168,169 t(:,kk+1)=0 0 0 0 1 1 0; case70,71,72,73,74,75,76,77,78,79,170,171,172,173,174,175,176,177,178,179 t(:,kk+1)=0 0 0 0 1 1 1; case80,81,82,83,84,85,86,87,88,89,180,181,182,183,184,185,186,187,188,189 t(:,kk+1)=0 0 0 1 0 0 0; case90,91,92,93,94,95,96,97,98,99,190,191,192,193,194,195,196,197,198,199 t(:,kk+1)=0 0 0 1 0 0 1; endendLOAD OK.save num_PT p t;% 创建和训练BP网络load num_PT p t;pr(1:512,1)=0;pr(1:512,2)=1;net=newff(pr,30 7,logsig logsig, traingdx, learngdm);% pr为输入节点 25 1隐层数目 。为隐层传输函数 net.trainParam.epochs=3000; %最大训练步数 %Backpropagation network training function %Backpropagation weight/bias learning function%net.trainParam.goal=0.0001; %训练目标误差net.tr