License plate recognition using OPENCV's SVM and neural network (high-quality article ******)

This article was reproduced from: http://blog.csdn.net/ap1005834/article/details/51340831

First, preface

This article refers to the automatic license plate recognition project in the deep understanding of OPENCV practical Computer Vision Project analysis, and then carries on the practice after understanding the method. The difficulty of accurately identifying the sequence of the characters in the license plate area is to realize the exact location of the license plate area. So the final recognition effect is still to be further improved.

second, the procedure flow

The program flow is as follows:

The corresponding main function is as follows

[CPP]  View plain  copy  print? #include   "Carid_detection.h"          int main ()    {            mat img_input = imread ("testcarid.jpg");        //If read-in image fails        if (Img_input.empty ())        {           fprintf (stderr,  "Can  not load image %s\n ", " testcarid.jpg ");            return -1;       }           Mat hsvImg ;       cvtcolor (IMG_INPUT,HSVIMG,CV_BGR2HSV);        vector<Mat> planes;       split ( Hsvimg,planes);   &NBSP;&NBSP;&NBSP;&NBSP;MAT&NBSP;SIMG;&NBSP;&NBSp     simg = planes[1];  //get red component         Blur (Simg,simg,size (3,3))    //3*3 Gaussian filter        vector < rotatedrect>  rects_simg;       posdetect (SIMG&NBSP;,RECTS_SIMG);                       mat grayimg;       rgbconvtogray (img_input ,grayimg);        medianblur (grayimg,grayimg,3);    //3*3 median filter            vector <RotatedRect>  rects_grayImg;        posdetect (grayimg ,rects_grayimg);          vector < ROTATEDRECT&GT;&NBSP;&NBSP;RECTS_CLOSEIMG;&NBSP;&NBSP;//license plate area closer to        posdetect_ Closeimg (SIMG&NBSP;,RECTS_CLOSEIMG);          vector <RotatedRect>   rects_optimal;       optimposdetect (rects_simg,rects_grayimg,rects_closeimg , rects_optimal);          vector <Mat> output_area;        normalposarea (Img_input ,rects_optimal,output_area);   // Get 144*33 's candidate license plate area output_area          CvSVM  svmClassifier;                        svm_train (svmclassifier)   //use SVM to train positive and negative samples             vector<mat> plates_svm;   //need to use the candidate plate area Output_area each pixel in the image as a line feature vector, and then make predictions.        for (Int i=0;i< output_area.size ();  ++i)        {           Mat img = output_area[i];            mat p = img.reshape (;  )         p.convertto (P,CV_32FC1);            int response =  (int) svmclassifier.predict ( p );