opencv3實現簡單的數字影象識別(KNN)
阿新 • • 發佈:2018-11-19
正在用opencv3做一個數字影象識別的小專案,要用到KNN,但是不熟悉它的介面,因此,借鑑了大佬的部落格,基本照搬了程式碼,程式碼如下:
大佬的連結如下:http://www.cnblogs.com/denny402/p/5033898.html
// knnrecognizenum.cpp:使用knn識別手寫數字 // #include "stdafx.h" #include<iostream> #include<opencv2\ml\ml.hpp> #include<highgui\highgui.hpp> using namespace std; using namespace cv; using namespace cv::ml; int main() { Mat img = imread("digits.png", 0); int boot = 20; int m = img.rows / boot; int n = img.cols / boot; Mat data, labels; //data和labels分別存放 //擷取資料的時候要按列擷取 for (int i = 0; i < n; i++) { int colNum = i * boot; for (int j = 0; j < m; j++) { int rowNum = j * boot; Mat tmp; img(Range(rowNum, rowNum + boot), Range(colNum, colNum + boot)).copyTo(tmp); data.push_back(tmp.reshape(0, 1)); //將影象轉成一維陣列插入到data矩陣中 labels.push_back((int)j / 5); //將影象對應的標註插入到labels矩陣中 } } data.convertTo(data, CV_32F); int sampleNum = data.rows; int trainNum = 3000; Mat trainData, trainLabel; trainData = data(Range(0, trainNum), Range::all()); trainLabel = labels(Range(0, trainNum), Range::all()); //使用KNN演算法 int k = 5; Ptr<TrainData> tData = TrainData::create(trainData,ROW_SAMPLE, trainLabel); //ROW_SAMPLE表示一行一個樣本 Ptr<KNearest> model = KNearest::create(); model->setDefaultK(k); model->setIsClassifier(true); model->train(tData); //預測分類 /* Mat sample = data.row(500); float res = model->predict(sample); cout << "預測結果是:"<< res << endl;*/ //預測一個的程式碼 double train_hr=0, test_hr=0; Mat response; for (int i = 0; i < sampleNum; i++) { Mat sample = data.row(i); float r = model->predict(sample); r = abs(r - labels.at<int>(i)); if (r <= FLT_EPSILON)// FLT_EPSILON表示最小的float浮點數,小於它,就是等於0 r = 1.f; else r = 0.f; if (i < trainNum) train_hr=train_hr+r; else test_hr=test_hr + r; } //cout << train_hr << " " << test_hr << endl; cout << "KNN模型在訓練集上的準確率為" << train_hr / trainNum * 100 << "%,在測試集上的準確率為" << test_hr / (data.rows-trainNum)*100<<"%"<<endl; system("pause"); return 0; }