之前说到，把车牌区域提前出来后，就可以着手识别程序了。先使用SVM判断是不是车牌。这里为了提高运行速度，板子资源有限，程序里我把svm训练部分注释掉了，假设每次都能找到车牌，实际使用时，还是要加上svm的。
然后对图像进行分割，我们的分类器只能对数字一个一个地识别，所以把每个数字分割出来，每个字符归一化为20*20的字符。
基本思想是先用findContours()函数把基本轮廓找出来，然后通过简单验证以确认是否为数字的轮廓。对于那些通过验证的轮廓，接下去会用boundingRect()找出它们的包围盒。
分割完后就可以进行识别了，字符识别使用ANN算法采用三层神经网络，识别需要用到一些xml文件，这些文件需要用分类器和大量样本做训练，提取他们的特征、，让机器去“学习”（利用训练好的XML文件去预测图像中车牌 ），我找的这三个xml数据集，说实话，不太好用，准确率一般般，有兴趣的可以自己训练。
完整程序如下，里面有详细注释了：
#include <opencv2/core/core.hpp>
#include <opencv2/highgui/highgui.hpp>
#include <opencv2/imgproc/imgproc.hpp>
#include <iostream>
#include <string>
#include <cvaux.h>
#include <stdio.h>
#include <opencv2/gpu/gpu.hpp>
#include <opencv2/ml/ml.hpp>
using namespace cv;
using namespace std;
//车牌宽高比为520/110=4.727272左右，误差不超过40%
//车牌高度范围在15~125之间，视摄像头距离而定（图像大小）
bool verifySizes_closeImg(const RotatedRect & candidate)
{
float error = 0.4;//误差40%
const float aspect = 4.7272;//44/14; //长宽比
int min = 15*aspect*15;//20*aspect*20; //面积下限，最小区域
int max = 125*aspect*125;//180*aspect*180; //面积上限，最大区域
float rmin = aspect - aspect*error; //考虑误差后的最小长宽比
float rmax = aspect + aspect*error; //考虑误差后的最大长宽比
int area = candidate.size.height * candidate.size.width;//计算面积
float r = (float)candidate.size.width/(float)candidate.size.height;//计算宽高比
if(r <1)
r = 1/r;
if( (area < min || area > max) || (r< rmin || r > rmax) )//满足条件才认为是车牌候选区域
return false;
else
return true;
}
void RgbConvToGray(const Mat& inputImage,Mat & outpuImage) //g = 0.3R+0.59G+0.11B
{
outpuImage = Mat(inputImage.rows ,inputImage.cols ,CV_8UC1);
for (int i = 0 ;i<inputImage.rows ;++ i)
{
uchar *ptrGray = outpuImage.ptr<uchar>(i);
const Vec3b * ptrRgb = inputImage.ptr<Vec3b>(i);
for (int j = 0 ;j<inputImage.cols ;++ j)
{
ptrGray[j] = 0.3*ptrRgb[j][2]+0.59*ptrRgb[j][1]+0.11*ptrRgb[j][0];
}
}
}
void normal_area(Mat &intputImg, vector<RotatedRect> &rects_optimal, vector <Mat>& output_area )
{
float r,angle;
for (int i = 0 ;i< rects_optimal.size() ; ++i)
{
//旋转区域
angle = rects_optimal[i].angle;
r = (float)rects_optimal[i].size.width / (float) (float)rects_optimal[i].size.height;
if(r<1)
angle = 90 + angle;//旋转图像使其得到长大于高度图像。
Mat rotmat = getRotationMatrix2D(rects_optimal[i].center , angle,1);//获得变形矩阵对象
Mat img_rotated;
warpAffine(intputImg ,img_rotated,rotmat, intputImg.size(),CV_INTER_CUBIC);
imwrite("car_rotated.jpg",img_rotated);//得到旋转图像
//裁剪图像
Size rect_size = rects_optimal[i].size;
if(r<1)
swap(rect_size.width, rect_size.height); //交换高和宽
Mat img_crop;
getRectSubPix(img_rotated ,rect_size,rects_optimal[i].center , img_crop );//图像切割
//用光照直方图调整所有裁剪得到的图像，使具有相同宽度和高度，适用于训练和分类
Mat resultResized;
//别人写的：
/*resultResized.create(33,144,CV32FC1);
resize(img_crop , resultResized,resultResized.size() , 0,0,INTER_CUBIC);
resultResized.convertTo(resultResized, CV32FC1);
resultResized = resultResized.reshape(1,1);*/
resultResized.create(33,144,CV_8UC3);//CV32FC1？？？？
resize(img_crop , resultResized,resultResized.size() , 0,0,INTER_CUBIC);
Mat grayResult;
RgbConvToGray(resultResized ,grayResult);
//blur(grayResult ,grayResult,Size(3,3));
equalizeHist(grayResult,grayResult);
output_area.push_back(grayResult);
}
}
bool char_verifySizes(const RotatedRect & candidate)
{
float aspect = 45.0f/77.0f;//45.0f/90.0f;
float width,height;
if (candidate.size.width >=candidate.size.height)
{
width = (float) candidate.size.height;
height = (float) candidate.size.width;
}
else
{
width = (float) candidate.size.width;
height = (float)candidate.size.height;
}
//这样确定是了高比宽要高
float charAspect = (float) width/ (float)height;//宽高比
float error = 0.35;//0.5;
float minHeight = 15; //最小高度11
float maxHeight = 28;//33; //最大高度33
float minAspect = 0.15;//0.05; //考虑到数字1，最小长宽比为0.15
float maxAspect = 1.0;
if( charAspect > minAspect && charAspect <= 1.0
&& height>= minHeight && height< maxHeight) //非0像素maxAspect长宽比、高度需满足条件
return true;
else
return false;
}
void char_sort(vector <RotatedRect > & in_char ) //对字符区域进行排序
{
vector <RotatedRect > out_char;
const int length = 7; //7个字符
int index[length] = {0,1,2,3,4,5,6};
float centerX[length];
for (int i=0;i < length ; ++ i)
{
centerX[i] = in_char[i].center.x;
}
for (int j=0;j <length;j++) {
for (int i=length-2;i >= j;i--)
if (centerX[i] > centerX[i+1])
{
float t=centerX[i];
centerX[i]=centerX[i+1];
centerX[i+1]=t;
int tt = index[i];
index[i] = index[i+1];
index[i+1] = tt;
}
}
for(int i=0;i<length ;i++)
out_char.push_back(in_char[(index[i])]);
in_char.clear(); //清空in_char
in_char = out_char; //将排序好的字符区域向量重新赋值给in_char
}
void char_segment(const Mat & inputImg,vector <Mat>& dst_mat)//得到20*20的标准字符分割图像
{
Mat img_threshold;
threshold(inputImg ,img_threshold , 180,255 ,CV_THRESH_BINARY );//二值化
//Mat element = getStructuringElement(MORPH_RECT ,Size(3 ,3)); //闭形态学的结构元素
//morphologyEx(img_threshold ,img_threshold,CV_MOP_CLOSE,element); //形态学处理
//imshow ("img_thresho00ld",img_threshold);
//waitKey();
Mat img_contours;
img_threshold.copyTo(img_contours);
if (!clearLiuDing(img_contours))
{
std::cout << "不是车牌" << endl;
}
else
{
//imshow("img_cda",img_contours);
//waitKey();
Mat result2;
inputImg.copyTo(result2);
vector < vector <Point> > contours;
findContours(img_contours ,contours,CV_RETR_EXTERNAL,CV_CHAIN_APPROX_NONE);
vector< vector <Point> > ::iterator itc = contours.begin();
vector<RotatedRect> char_rects;
drawContours(result2,contours,-1, Scalar(0,255,255), 1);
while( itc != contours.end())
{
RotatedRect minArea = minAreaRect(Mat( *itc )); //返回每个轮廓的最小有界矩形区域
Point2f vertices[4];
minArea.points(vertices);
if(!char_verifySizes(minArea)) //判断矩形轮廓是否符合要求
{
itc = contours.erase(itc);
}
else
{
++itc;
char_rects.push_back(minArea);
}
}
/*imshow("char1",char_rects[1]);
imshow("char2",char_rects[2]);
imshow("char3",char_rects[3]);
imshow("char4",char_rects[4]);
imshow("char5",char_rects[5]);
imshow("char6",char_rects[6]);
imshow("char7",char_rects[0]);
waitKey();*/
char_sort(char_rects); //对字符排序
vector <Mat> char_mat;
for (int i = 0; i<char_rects.size() ;i++ )
{
char_mat.push_back(Mat(img_threshold,char_rects[i].boundingRect()));
}
//imshow("char_mat1",char_mat[0]);
//imshow("char_mat2",char_mat[1]);
//imshow("char_mat3",char_mat[2]);
//imshow("char_mat4",char_mat[3]);
//imshow("char_mat5",char_mat[4]);
//imshow("char_mat6",char_mat[5]);
//imshow("char_mat7",char_mat[6]);
//waitKey();
Mat train_mat(2,3,CV_32FC1);
int length ;
dst_mat.resize(7);
Point2f srcTri[3];
Point2f dstTri[3];
for (int i = 0; i==0;i++)
{
srcTri[0] = Point2f( 0,0 );
srcTri[1] = Point2f( char_mat[i].cols - 1, 0 );
srcTri[2] = Point2f( 0, char_mat[i].rows - 1 );
length = char_mat[i].rows > char_mat[i].cols?char_mat[i].rows:char_mat[i].cols;
dstTri[0] = Point2f( 0.0, 0.0 );
dstTri[1] = Point2f( length, 0.0 );
dstTri[2] = Point2f( 0.0, length );
train_mat = getAffineTransform( srcTri, dstTri );
dst_mat[i]=Mat::zeros(length,length,char_mat[i].type());
warpAffine(char_mat[i],dst_mat[i],train_mat,dst_mat[i].size(),INTER_LINEAR,BORDER_CONSTANT,Scalar(0));
//resize(dst_mat[i],dst_mat[i],Size(20,20),0,0,CV_INTER_CUBIC); //尺寸调整为20*20
resize(dst_mat[i],dst_mat[i],Size(20,20));//每个字符归一化为20*20的字符
}
for (int i = 1; i< char_mat.size();++i)
{
srcTri[0] = Point2f( 0,0 );
srcTri[1] = Point2f( char_mat[i].cols - 1, 0 );
srcTri[2] = Point2f( 0, char_mat[i].rows - 1 );
length = char_mat[i].rows > char_mat[i].cols?char_mat[i].rows:char_mat[i].cols;
dstTri[0] = Point2f( 0.0, 0.0 );
dstTri[1] = Point2f( length, 0.0 );
dstTri[2] = Point2f( 0.0, length );
train_mat = getAffineTransform( srcTri, dstTri );
dst_mat[i]=Mat::zeros(length,length,char_mat[i].type());
warpAffine(char_mat[i],dst_mat[i],train_mat,dst_mat[i].size(),INTER_LINEAR,BORDER_CONSTANT,Scalar(0));
//resize(dst_mat[i],dst_mat[i],Size(20,20),0,0,CV_INTER_CUBIC); //尺寸调整为20*20
resize(dst_mat[i],dst_mat[i],Size(20,20));//每个字符归一化为20*20的字符
}
}
}
void features(const Mat & in , Mat & out ,int sizeData)
{
// 分别在水平方向和垂直方向上 创建累积直方图
Mat vhist = projectHistogram(in , 1); //水平直方图
Mat hhist = projectHistogram(in , 0); //垂直直方图
// 低分辨率图像
// 低分辨率图像中的每一个像素都将被保存在特征矩阵中
Mat lowData;
resize(in , lowData ,Size(sizeData ,sizeData ));
//特征矩阵的列数
int numCols = vhist.cols + hhist.cols + lowData.cols * lowData.cols;
out = Mat::zeros(1, numCols , CV_32F);
// 向特征矩阵赋值
int j = 0;
for (int i =0 ;i<vhist.cols ; ++i)// 首先把水平方向累积直方图的值，存到特征矩阵中
{
out.at<float>(j) = vhist.at<float>(i);
j++;
}
for (int i=0 ; i < hhist.cols ;++i)// 然后把竖直方向累积直方图的值，存到特征矩阵中
{
out.at<float>(j) = hhist.at<float>(i);
}
for(int x =0 ;x<lowData.rows ;++x)// 最后把低分辨率图像的像素值，存到特征矩阵中
{
for (int y =0 ;y < lowData.cols ;++ y)
{
out.at<float>(j) = (float)lowData.at<unsigned char>(x,y);
j++;
}
}
}
void ann_train(CvANN_MLP &ann ,int numCharacters, int nlayers, string str)//http://blog.csdn.net/yiqiudream/article/details/51712497
{
Mat trainData ,classes;
FileStorage fs;
fs.open(str, FileStorage::READ);//str是文件名字
fs["TrainingData"] >>trainData;
fs["classes"] >>classes;
//CvANN_MLP bp;
//Set up BPNetwork's parameters
//CvANN_MLP_TrainParams params;
//params.train_method=CvANN_MLP_TrainParams::BACKPROP;
//params.bp_dw_scale=0.1;
//params.bp_moment_scale=0.1;
Mat layerSizes(1,3,CV_32SC1);
layerSizes.at<int>( 0 ) = trainData.cols;
layerSizes.at<int>( 1 ) = nlayers; //隐藏神经元数，可设为3
layerSizes.at<int>( 2 ) = numCharacters; //样本类数为34
//layerSizes.at<int>( 3 ) = numCharacters ;
ann.create(layerSizes , CvANN_MLP::SIGMOID_SYM ); //初始化ann
Mat trainClasses;
trainClasses.create(trainData.rows , numCharacters ,CV_32FC1);
for (int i =0;i< trainData.rows; i++)
{
for (int k=0 ; k< trainClasses.cols ; k++ )
{
if ( k == (int)classes.at<uchar> (i))
{
trainClasses.at<float>(i,k) = 1 ;
}
else
trainClasses.at<float>(i,k) = 0;
}
}
Mat weights(1 , trainData.rows , CV_32FC1 ,Scalar::all(1) );
ann.train( trainData ,trainClasses , weights);
}
void svm_train(CvSVM & svmClassifier)
{
FileStorage fs;
fs.open("SVM.xml" , FileStorage::READ);
Mat SVM_TrainningData;
Mat SVM_Classes;
fs["TrainingData"] >>SVM_TrainningData;
fs["classes"] >>SVM_Classes;
CvSVMParams SVM_params;
SVM_params.kernel_type = CvSVM::LINEAR;
svmClassifier.train(SVM_TrainningData,SVM_Classes ,Mat(),Mat(),SVM_params); //SVM训练模型
fs.release();
}
int main(int argc, char* argv[])
{
Mat img_input= imread("./car.jpg");//加载图片
if(img_input.empty())//如果读入图像失败
{
cout << "Can not load image" << endl;
return -1;
}
Mat hsvImg ;
cvtColor(img_input,hsvImg,CV_BGR2HSV);//RGB模型转换成HSV模型
imwrite("car_hsv.jpg",hsvImg);//看下hsv效果
vector <Mat> hsvSplit;
split(hsvImg,hsvSplit);//将图像的各个通道分离
equalizeHist(hsvSplit[2],hsvSplit[2]);//直方图均衡化,提高图像的质量
merge(hsvSplit,hsvImg);//将分离的多个单通道合成一幅多通道图像
imwrite("car_hsv1.jpg",hsvImg);//看下处理效果
const int min_blue =100;//最小蓝车区域
const int max_blue =240;//最大蓝车区域
int avg_h = (min_blue+max_blue)/2;
int channels = hsvImg.channels();
int nRows = hsvImg.rows;
//图像数据列需要考虑通道数的影响；
int nCols = hsvImg.cols * channels;
if (hsvImg.isContinuous())//连续存储的数据，按一行处理
{
nCols *= nRows;
nRows = 1;
}
int i, j;
unsigned char* p;
const float minref_sv = 64; //参考的S V的值
const float max_sv = 255; // S V 的最大值
for (i = 0; i < nRows; ++i)//根据蓝色在HSV在的区域每个通道的取值范围将此作为阈值，提取出图片中蓝色部分作为备选区域
{
p = hsvImg.ptr<uchar>(i);//有效提高了车牌和车色颜色在不相差较大的情况下的识别率
for (j = 0; j < nCols; j += 3)//致命问题:蓝色的车和蓝色的牌照?
{
int H = int(p[j]); //0-180
int S = int(p[j + 1]); //0-255
int V = int(p[j + 2]); //0-255
bool colorMatched = false;
if (H > min_blue && H < max_blue)
{
int Hdiff = 0;
float Hdiff_p = float(Hdiff) / 40;
float min_sv = 0;
if (H > avg_h)
{
Hdiff = H - avg_h;
}
else
{
Hdiff = avg_h - H;
}
min_sv = minref_sv - minref_sv / 2 * (1 - Hdiff_p);
if ((S > 70&& S < 255) &&(V > 70 && V < 255))
colorMatched = true;
}
if (colorMatched == true)
{
p[j] = 0; p[j + 1] = 0; p[j + 2] = 255;
}
else
{
p[j] = 0; p[j + 1] = 0; p[j + 2] = 0;
}
}
}
Mat src_grey;
Mat img_threshold;
vector<Mat> hsvSplit_done;
split(hsvImg, hsvSplit_done);
src_grey = hsvSplit_done[2];//提取黑色分量
imwrite("car_hsvSplit.jpg",src_grey);//查看分离通道出来的车牌
vector <RotatedRect> rects;
Mat element = getStructuringElement(MORPH_RECT ,Size(17 ,3)); //闭形态学的结构元素
morphologyEx(src_grey ,img_threshold,CV_MOP_CLOSE,element); //闭运算，先膨胀后腐蚀，连通近邻区域（填补白色区域的间隙）
morphologyEx(img_threshold,img_threshold,MORPH_OPEN,element);//形态学处理
imwrite("car_morphology.jpg",img_threshold);//查看threshold
vector< vector <Point> > contours;//寻找车牌区域的轮廓
findContours(img_threshold ,contours,CV_RETR_EXTERNAL, CV_CHAIN_APPROX_NONE);//只检测外轮廓。存储所以轮廓点
//绘制轮廓
/*for(int find=0; find < contours.size(); find++)
drawContours(img_threshold, contours, find, Scalar(255), 2);
imwrite("car_contours.jpg",img_threshold);//查看轮廓*/
//对候选的轮廓进行进一步筛选
vector< vector <Point> > ::iterator itc = contours.begin();
while( itc != contours.end())
{
RotatedRect mr = minAreaRect(Mat( *itc )); //返回每个轮廓的最小有界矩形区域
if(!verifySizes_closeImg(mr)) //判断矩形轮廓是否符合要求
{
itc = contours.erase(itc);
}
else
{
rects.push_back(mr);
++itc;
}
}
vector <Mat> output_area;
normal_area(img_input ,rects,output_area); //获得144*33的候选车牌区域output_area
imwrite("car_area.jpg",output_area[0]);//得到候选区域，这里可能会获得多个候选区域，最好使用svm训练一下
//（二）添加如下：
//CvSVM svmClassifier;//为了运行速度，我就把这里注释掉了，这样会降低准确度
//svm_train(svmClassifier); //使用SVM对正负样本进行训练，为了运行速度，我就把这里注释掉了，这样会降低准确度
vector<Mat> plates_svm; //需要把候选车牌区域output_area图像中每个像素点作为一行特征向量，后进行预测
for(int i=0;i< output_area.size(); ++i)//实际情况下应该加上SVM训练，我这里是学习测试
{
cout << "output " << i << endl;
Mat img = output_area[i];
Mat p = img.reshape(1,1);
p.convertTo(p,CV_32FC1);
//int response = (int)svmClassifier.predict( p );//为了运行速度，我就把这里注释掉了，这样会降低准确度
//if (response == 1)//为了运行速度，我就把这里注释掉了，这样会降低准确度
plates_svm.push_back(output_area[i]); //保存预测结果
}
//从SVM预测获取车牌区域分割得到字符区域
vector <Mat> char_seg;
char_segment(plates_svm[0],char_seg);//对车牌区域中字符进行分割
imwrite("char0.jpg",char_seg[0]);//显示七个字符
imwrite("char1.jpg",char_seg[1]);
imwrite("char2.jpg",char_seg[2]);
imwrite("char3.jpg",char_seg[3]);
imwrite("char4.jpg",char_seg[4]);
imwrite("char5.jpg",char_seg[5]);
imwrite("char6.jpg",char_seg[6]);
//获得7个字符矩阵的相应特征矩阵
vector <Mat> char_feature;
char_feature.resize(7);
for (int i =0;i<char_seg.size() ;++ i)
features(char_seg[i], char_feature[i],5);
//神经网络训练
CvANN_MLP ann_classify;//对字母和数字
ann_train(ann_classify,34,20,"ann_xml.xml");//输入层经元数(离线训练数据集的行数)，隐藏层的神经元数，文件名字
CvANN_MLP ann_classify1;//对第一个汉字进行分类建模
ann_train(ann_classify1,3,20,"ann_xml_character.xml");
//字符预测
vector<int> char_result;
//classify(ann_classify,char_feature,char_result);
char_result.resize(char_feature.size());
for (int i=0;i<char_feature.size(); ++i)
{
if (i==0)//对汉字
{
Mat output(1 ,34, CV_32FC1); //1*34矩阵
//ann.predict(char_feature[i] ,output);
ann_classify1.predict(char_feature[i],output);//对每个字符运用ANN.predict函数得出1*类别数的数据组（数据组中是记录这个字符跟每个类别的“相似度”）
Point maxLoc;
double maxVal;
minMaxLoc(output , 0 ,&maxVal , 0 ,&maxLoc);//找出最大概率的类别
char_result[i] = maxLoc.x;
}
else//对字母和数字
{
Mat output(1 ,34, CV_32FC1); //1*34矩阵
//ann.predict(char_feature[i] ,output);
ann_classify.predict(char_feature[i],output);//预测
Point maxLoc;
double maxVal;
minMaxLoc(output , 0 ,&maxVal , 0 ,&maxLoc);
char_result[i] = maxLoc.x;
}
}
if(plates_svm.size() != 0)
{
cout << "create a image" << endl;
imwrite("car_opencv_final.jpg",output_area[0]); //正确预测的话，就只有一个结果plates_svm[0]
}
else
{
std::cout<<"定位失败";
return -1;
}
cout<<"该车牌后7位为：";
char s[] = {'0','1','2','3','4','5','6','7','8','9','A','B',
'C','D','E','F','G','H','J','K','L','M','N','P','Q',
'R','S','T','U','V','W','X','Y','Z'};//现在添加了京
cout<<'\n';
string chinese[]={"湘","鄂","粤","甘","贵","桂","黑","沪",
"冀","津","京","吉","辽","鲁","蒙","闵","宁","青","琼",
"陕","苏","晋","皖","湘","浙","豫","渝","粤","云"};
for (int w=0;w<char_result.size(); w++) //第一位是汉字，这里没实现对汉字的预测
{
if (w==0)
{
cout<<chinese[char_result[w]];//可以对汉字京的识别
cout<<'\t';
}
else
{
cout<< s[char_result[w]];
cout<<'\t';
}
}
return 0;
}
源代码和xml文件下载：https://download.csdn.net/download/guet_kite/10930196
代码大部分都是抄网上一篇文章的，文章地址我找不到了…
不过给出几个参考链接，可以看看里面的。
参考：
[1]:使用opencv的SVM和神经网络实现车牌识别
[2]:OpenCV自学笔记17. 基于SVM和神经网络的车牌识别
[3]:OpenCV实现车牌识别，OCR分割，ANN神经网络