svm

gistfile1.txt

#include <opencv2/core/core.hpp>
#include <opencv2/highgui/highgui.hpp>
#include <opencv2/imgproc/imgproc.hpp>
#include <opencv2/ml/ml.hpp>
#include <iostream>
#include <stdio.h>
#include <stdlib.h>
#include <string>
#include <vector>
#include <algorithm>
using namespace cv;
using namespace std;

class TrainingSet {
    
public:
    Mat data;
    int label;
    TrainingSet(Mat data, int label) {
        this->data = data;
        this->label = label;
    }

    
};

int writeTrainingSet() {
    Mat image;
    int fileName;
    String fullFileName;
    vector<Mat> trainingData;
    vector<int> label;
    
    Mat train;
    
    for(int folderName=0;folderName<=9;folderName++){
        fileName = 1;
        
        while(true){
            fullFileName = "/Users/nuttt/MNIST/train/"+to_string(folderName)+"/"+to_string(fileName)+".jpg";
            image = imread(fullFileName, CV_LOAD_IMAGE_GRAYSCALE);
            cout << fullFileName << endl;
            
            if(!image.data){
                cout <<  "Could not open or find the image" << std::endl ;
                break;
            }
            Mat tmp = image.reshape(0,1);
            trainingData.push_back(tmp);
            label.push_back(folderName);
            
            fileName++;
//            break;
        }
//        break;
    }
    
    
    

    
    FileStorage fs("/Users/nuttt/MNIST/train.yml.gz", FileStorage::WRITE);
    fs << "trainingData" << trainingData;
    fs << "trainingLabel" << label;
    fs.release();
    
    return 0;
}




int writeTestSet() {
    Mat image;
    int fileName;
    String fullFileName;
    vector<Mat> trainingData;
    vector<int> label;
    
    for(int folderName=0;folderName<=9;folderName++){
        fileName = 1;
        
        while(true){
            fullFileName = "/Users/nuttt/MNIST/test/"+to_string(folderName)+"/"+to_string(fileName)+".jpg";
            image = imread(fullFileName, CV_LOAD_IMAGE_GRAYSCALE);
            cout << fullFileName << endl;
            
            if(!image.data){
                cout <<  "Could not open or find the image" << std::endl ;
                break;
            }
            trainingData.push_back(image.reshape(0,1));
            label.push_back(folderName);
            
            fileName++;
            //            break;
        }
        //        break;
    }
    FileStorage fs("/Users/nuttt/MNIST/test.yml.gz", FileStorage::WRITE);
    fs << "testData" << trainingData;
    fs << "testLabel" << label;
    fs.release();
    
    return 0;
}


vector<TrainingSet> readTrainingSet() {
    FileStorage fs("/Users/nuttt/MNIST/train.yml.gz", FileStorage::READ);
    vector<TrainingSet> output;
    vector<Mat> data;
    vector<int> label;
    
    fs["trainingData"] >> data;
    fs["trainingLabel"] >> label;
    
    for(int i=0; i<label.size(); i++){
        output.push_back(TrainingSet(data[i], label[i]));
    }
    
    fs.release();
    
    return output;
}

vector<TrainingSet> readTestSet() {
    FileStorage fs("/Users/nuttt/MNIST/test.yml.gz", FileStorage::READ);
    vector<TrainingSet> output;
    vector<Mat> data;
    vector<int> label;
    
    fs["testData"] >> data;
    fs["testLabel"] >> label;
    
    for(int i=0; i<label.size(); i++){
        output.push_back(TrainingSet(data[i], label[i]));
    }
    
    fs.release();
    
    return output;
}

int predict(vector<TrainingSet> trainingData, Mat testData){
    vector<pair<double, int> > distance;
    int count[10] = {0,0,0,0,0,0,0,0,0,0};
    double d;
    Mat test = testData.reshape(0, 1);
    for(int i=0; i<trainingData.size(); i++){
        d = norm(trainingData[i].data, test);
        distance.push_back(pair<double, int> (d, trainingData[i].label));
    }
    sort(distance.begin(), distance.end());
    for(int i=0; i<10; i++){
        count[distance[i].second]++;
    }
    int maxx = count[0];
    int maxID = 0;
    for(int i=0; i<10; i++){
        if(count[i] > maxx){
            maxx = count[i];
            maxID = i;
        }
    }
    
    return maxID;
}

void trainSVM() {
    vector<TrainingSet> trainingData;
    trainingData = readTrainingSet();
    
    CvSVM svm;
    
    Mat train, label;
    
    for(int i = 0; i < trainingData.size(); i++) {
        train.push_back(trainingData[i].data);
        label.push_back(trainingData[i].label);
    }
    
    train.convertTo(train, CV_32F);
    
    CvSVMParams params;
    
    params.svm_type = CvSVM::C_SVC;
    params.kernel_type = CvSVM::LINEAR;
    
    cout << "Training SVM" << endl;
    
    svm.train(train, label, Mat(), Mat(), params);
    
    svm.save("/Users/nuttt/MNIST/svm_linear.gz");
}


int main(int, char**) {
    
    cout << "Start" << endl;
    
    //trainSVM();
    
    
    vector<TrainingSet> testData;
    testData = readTestSet();
    
    CvSVM svm;
    
    svm.load("/Users/nuttt/MNIST/svm_linear.gz");
    
    Mat result, input;
    
    int result_count[10][10]; //[actual][predict]
    
    for (int i=0; i<10; i++) {
        for (int j=0; j<10; j++) {
            result_count[i][j] = 0;
        }
    }
    
    for(int i=0; i<testData.size();i++) {
        
        testData[i].data.convertTo(input, CV_32F);
        int response = svm.predict(input);
        
        cout << testData[i].label << ": " << response << endl;
        
        result_count[response][testData[i].label]++;
    }
    
    for (int i=0; i<10; i++) {
        for (int j=0; j<10; j++) {
            printf("%d\t", result_count[i][j]);
        }
        printf("\n");
    }

    
    return 0;
}