《ofo车牌识别》研发心得06-OpenCV的SVM

图像预处理搞定，可以扔进SVM做训练和预测了。

SVM相当简单，但要万分注意OpenCV的SVM要CV_32S

训练

//转成Mat
Mat trainningDataMat(countLines,32/8*72/8,CV_32FC1,trainingData);
Mat labelsMat(countLines,1,CV_32S,labels); //注意label要用CV_32S，不能用CV_32FC1。否则会报错

//训练SVM
Ptr<SVM> svm = SVM::create();
svm->setType(SVM::C_SVC);
svm->setKernel(SVM::POLY);
svm->setDegree(3);
svm->setGamma(0.01);
svm->setCoef0(2);
svm->setTermCriteria(TermCriteria(TermCriteria::MAX_ITER, 100, 1e-6));
svm->train(trainningDataMat,ROW_SAMPLE, labelsMat);
svm->save("svmtrain.txt");

预测(注意预测要逐行逐行来)

//转成Mat(预测只能逐行逐行来)
for (int i=0;i<countLines;i++)
{
    Mat trainningDataMat(1,32/8*72/8,CV_32FC1,trainingData[i]);
    //预测SVM
    cv::Ptr<cv::ml::SVM> mSvm2;
    mSvm2 = cv::ml::SVM::load<cv::ml::SVM>("svmtrain.txt");
    float q = mSvm2->predict(trainningDataMat);
    cout<<"predict: "<<q<<"\treal:"<<labels[i]<<endl;
    predictData<<"predict: "<<q<<"\treal:"<<labels[i]<<endl;
}

成功！Yeah！