gistfile1.txt

#include "opencv2/opencv.hpp"
using namespace cv;

#include <iostream>
using namespace std;

//
// b g r y cr cb h s v like pearls ona string.
//
Mat feature(const Vec3b &pixel) {
    // normalized bgr;
    float b = float(pixel[0]) / 255;
    float g = float(pixel[1]) / 255;
    float r = float(pixel[2]) / 255;

    // ycrcb
    float y  = 0.299f * r + 0.587f * g + 0.114f * b;
    float cr = 0.713f * (r - y) + 0.5f;
    float cb = 0.564f * (b - y) + 0.5f;

    // hsv
    float M(max(max(b,g),r));
    float m(min(min(b,g),r));
    float md = M - m;
    float A = 1.0f / 6;
    float v =  M;
    float s = (M>0)   ? (md/M) : 0;
    float h = (md==0) ? 0 :
              (M==r)  ? (    A*(g-b)/md) :
              (M==g)  ? (2*A+A*(b-r)/md) :
                        (3*A+A*(r-g)/md);

    return Mat_<float>(1,9) << b,g,r, y,cr,cb, h,s,v; // 9x1
}

//
// txt file with 3 numbers on a row, b g r
// (made a little colorpicker, that samples 8nb per click)
int getPos(Mat &features, int off, int n)
{
    ifstream in("skin_bgr.txt");
    int b=0,g=0,r=0, i=0;
    while ((in >> b) && (in >> g) && (in >> b))
    {
        i++;
        if (i<=off) continue;
        if (i>off+n) break;
        Mat row = feature(Vec3b(b,g,r));
        features.push_back(row);
    }
    return features.rows;
}

//
// negative images from browser cache, no humans allowed !
// (a whole hardcoded folder)
int getNeg(Mat &features, int off, int n)
{
    int k=0, of=features.rows;
    vector<String> fn;
    glob("c:/data/img/noskin", fn);
    for (size_t i=0; i<fn.size(); i++)
    {
        Mat img = imread(fn[i]);
        if (img.empty())
            continue;
        for (int i=0; i<200; i++) // N random pixel samples per image
        {
            k++;
            if (k<=off) continue;
            if (k>off+n) break;

            int x = theRNG().uniform(5,img.cols-5);
            int y = theRNG().uniform(5,img.rows-5);
            Vec3b p = img.at<Vec3b>(y,x);
            Mat row = feature(p);
            features.push_back(row);
        }
    }
    return features.rows - of;
}

int main()
{
    int ntest = 1000;
    //print_colors();
    Mat features;
    int p = getPos(features, ntest, 0xffff); // save N for testing later
    cout << p << " positive, ";

    int n = getNeg(features, ntest, 0xffff);
    cout << n << " negative probes, ";
    cout << features.size() << " feature elems." << endl;

    Mat labels(features.rows, 1, CV_32S, Scalar(1));
    labels(Range(p,labels.rows),Range::all()) = 0;

    Ptr<ml::SVM> svm = ml::SVM::create();
    svm->setType(ml::SVM::NU_SVC);
    svm->setKernel(ml::SVM::LINEAR);
    svm->setNu(0.55f);
    svm->setTermCriteria(TermCriteria(TermCriteria::MAX_ITER+TermCriteria::EPS, 1000, 1e-6));
    svm->setC(60);

    bool ok = svm->train(features , ml::ROW_SAMPLE , labels);
    cout << "training done: " << ok << "." << endl;
    svm->save("colors_svm.xml");
    Mat query;
    p = getPos(query, 0, ntest); // first N are for testing
    cout << p << " positive, ";

    n = getNeg(query, 0, ntest);
    cout << n << " negative probes, ";
    cout << query.size() << " query elems." << endl;

    Mat truth(query.rows, 1, CV_32F, Scalar(1));
    truth(Range(p,truth.rows),Range::all()) = 0;

    Ptr<ml::SVM> svm2 = ml::SVM::load("colors_svm.xml");

    Mat pred;
    svm2->predict(query, pred);

    Mat_<int> confusion(2,2, 0);
    for (size_t i=0; i<pred.total(); i++)
    {
        int r = int(pred.at<float>(i));
        int c = int(truth.at<float>(i));
        confusion(r, c) ++;
    }
    cout << confusion << endl;

    /* Mat eq(pred == truth);
    int nz = countNonZero(eq);
    int nelm = (ntest*2); */
    int nz = sum(confusion.diag())[0];
    int nelm = sum(confusion)[0];
    float acc =  (float(nz) / nelm);
    cout << nz << "/" << nelm << " : " << acc << endl;
    return 0;
}