bob409's profile - activity

I have used HOGDescriptors to compute the feature vector of an image, now I have to use SVM to create training set for a class. I have to create multiclass SVM as i have 6 classes each with 10 images. I do not know how to convert vector into Mat and how to label the class to create an xml file?

hog.compute(drawing, ders, Size(1, 1), Size(0, 0));



Mat Hogfeat;
Hogfeat.create(ders.size(), 1, CV_32FC1);

for (int i = 0; i<ders.size(); i++)
{
    Hogfeat.at<float>(i, 0) = ders.at(i);


}

int labels = {1};
Mat labelsMat(1, 1, CV_32SC1, labels);

Ptr<ml::SVM> svm = ml::SVM::create();
svm->setType(ml::SVM::C_SVC);
svm->setKernel(ml::SVM::LINEAR);
svm->setGamma(3);
svm->setTermCriteria(TermCriteria(TermCriteria::MAX_ITER, 100, 1e-6));

Ptr<ml::TrainData> tData = ml::TrainData::create(Hogfeat, ml::SampleTypes::ROW_SAMPLE, labelsMat);
svm->train(tData);

svm->save("testing.xml");

I am storing features in form of a matrix in an xml file using FileStorage

FileStorage f("trial.xml",FileStorage::WRITE);
Mat feature;
f<<"feature"<<feature;

By executing the above code, a trial.xml file is created

<?xml version="1.0"?>
<opencv_storage>
<features type_id="opencv-matrix">
  <rows>1</rows>
  <cols>8</cols>
  <dt>f</dt>
  <data>
   3.35365832e-001 1.21951215e-002 9.14634094e-002 6.30081296e-002
    3.31300795e-001 1.01626012e-002 9.95934904e-002 5.69105670e-002</data></features>
 </opencv_storage>

However, I cannot retrive the matrix from the xml file.

FileStorage fs("trial.xml",FileStorage::READ);
Mat data;
fs["data"]>>data;
cout<<data<<endl;

The Mat data is empty: []

I have not thrown some random numbers. I am working with KNN and I got this problem that is why I am asking. I do not understand how to get the correct label. Thanks for the link. It is the sample which which I have trained the data so knn should output label 1

I have alreday read that. What I am asking is that is class 1 dist is 0 why am i getting response as class 0 instead of class 1

I have trained KNN to recognise obj A as 1

When I am doinh knn->findKnearest(data,m,response,neighbor,dist);

I get response as 0; Then neighbor [1,0,...] dist[0,0.23333]

What does this mean? Do I have to use euclidean distance to do matching?

How to input labels for classes in NN. Are they like SVM give different labels to each class. I have not really understood how to create the training class for neural network? I do not know how to use it for training I have tried the following way:

void mlp(cv::Mat& trainingData, vector<int>& index)
{
    int input_neurons = 8;
    int hidden_neurons = 100;
    int output_neurons = 12;
    Mat layerSizes = Mat(3, 1, CV_32SC1);
    layerSizes.row(0) = Scalar(input_neurons);
    layerSizes.row(1) = Scalar(hidden_neurons);
    layerSizes.row(2) = Scalar(output_neurons);

    Ptr<ml::ANN_MLP> mlp = ml::ANN_MLP::create();
    mlp->setLayerSizes(layerSizes);
    mlp->setTrainMethod(ml::ANN_MLP::SIGMOID_SYM);
    mlp->setTermCriteria(TermCriteria(CV_TERMCRIT_ITER | CV_TERMCRIT_EPS, 1000, 0.00001f));
    mlp->setTrainMethod(ml::ANN_MLP::BACKPROP,0.1f,0.1f);
    mlp->setActivationFunction(ml::ANN_MLP::SIGMOID_SYM, 1, 1);

    Mat trainClasses;

    cout << "Poker" << endl;
    trainClasses.create(trainingData.rows, 12, CV_32FC1);
    for (int i = 0; i < trainClasses.rows; i++)
    {
        trainClasses.at<float>(i, index[i]) = 1.f;
    }

    cout << "Row of trainClass: " << trainClasses.cols << endl;
    cout << "Row of trainData: " << trainingData.cols << endl;
    cout << "Koker" << endl;
    Ptr<ml::TrainData> td = ml::TrainData::create(trainingData, ml::ROW_SAMPLE, trainClasses);
    mlp->train(td);
    cout << "Training Done" << endl;

    mlp->save("neural_network.xml");

}

When I am predicting, I am getting the following response:

[-4.4227011e+0008,-404295997e+008,-4.4214237e+008,-4.3398125e+008,.......]

What is wrong and how to interpret the response from an ANN?

I am getting this error:

OpenCV Error: Assertation failed <samples.type<>CV_32F || samples.type<>CV_32S> in cv::ml::TrainDataImpl::setData

I have extracted feature and placed it in a String;

//for loop

String features = "352111653512146120635735005";

Mat m;
    m.push_back(features);
    m.reshape(1, 1);

    freeman.push_back(m);
             labelMat.push_back(label);

             Ptr<ml::KNearest> knn = ml::KNearest::create();
        Ptr<ml::TrainData> td = ml::TrainData::create(feature, ml::ROW_SAMPLE, label);  <-----I am getting error here

What is the problem in doing that if I am getting the result? I know only c-Api to extract chain code.

I have extracted two feature vectors from an image:

1. chain code histogram
2. fourier descriptors

How to fuse them and train them using svm?

I have tried hconcat but it is not working. When i am concatenatenating: hconcat(fourierdescriptor,mergeMatrix)

mergeMatrix is Null

 Mat fd;

vector<vector<Point> > contours;
vector<Vec4i> hierarchy;
findContours(inputImage, contours, hierarchy, CV_RETR_TREE, CV_CHAIN_APPROX_SIMPLE, Point(0, 0));

Mat drawing = Mat::zeros(inputImage.size(), CV_8UC1);

int s = findBiggestContour(contours);

drawContours(drawing, contours, s, Scalar(255), 1, 8, hierarchy, 0, Point());

imshow("draw", drawing);

vector<int> ctrSelec;
ctrSelec.push_back(s);

vector<Point2d> z;
vector<Point2d> Z;

z.resize(ctrSelec.size());
Z.resize(ctrSelec.size());


vector<Point2d> c = ReSampleContour(contours[ctrSelec[0]], 30);

Mat mc = Mat::zeros(drawing.size(), CV_8UC3);


    for (int j = 0; j < c.size(); j++)
    {
        z.push_back(c[(j * 10) % c.size()]);

        dft(z, Z, DFT_SCALE | DFT_REAL_OUTPUT);
    }


    fd.push_back(Z);

    for (int i = 0; i < Z.size(); i++)
    {
        cout << Z[i].x << "," << Z[i].y << endl;
    }
//medianBlur(drawing, drawing, 3);

CvChain* chain;
CvMemStorage* storage = 0;
storage = cvCreateMemStorage();
cvFindContours(&IplImage(drawing), storage, (CvSeq**)(&chain), sizeof(*chain), CV_RETR_EXTERNAL, CV_CHAIN_CODE);



int total = chain->total;

cv::Mat hist(1, 8, CV_32F, Scalar(0));

int totalCount = 0;
for (; chain != NULL; chain = (CvChain*)chain->h_next)
{

    int numChain = 0;

    CvSeqReader reader;
    int i, total = chain->total;

    cvStartReadSeq((CvSeq*)chain, &reader, 0);


    for (i = 0; i < total; i++)
    {
        char code;
        CV_READ_SEQ_ELEM(code, reader);
        int Fchain = (int)code;


        hist.at<float>(0, Fchain)++;

        totalCount++;
    }
}


    Mat prob = hist / totalCount;


    cch = prob.clone();

    Mat feature;
           hconcat(fd,cch,feature)

So was it good?

I dnt nid fourier descriptor which are rotational invariant they just need to be translation and scale invariant. It means that u have 1024 fourier descriptors

What should be taken into consideration while choosing a value? Why have u chosen 1024, was it derived by experimental result? And where do u get the fourier descriptors in the code above? Have u made them scale and starting point invariant? If yes where?

What does the following line perform? vector<point2d> c = ReSampleContour(contours[ctrSelec[i]],1024); Why 1024? Has anyone understood it?

When using binary svm for classification, it returns a value when using a value when using predict? What does the value mean? It is the distance between the test data and the predicted data. If someone knows the answer please help.

What is not working> What error are you getting?

How to normalize hog feature?

You have to set environmental variables.Then you have to change some setting in ur project->properties. This video will help u https://www.youtube.com/watch?v=JrIpA.... Just make the necessary changes as in the video the person is using vs13 and opencv 3.0 n u vs10 with opencv 3.1.0. Try it

Thnks. Should I normalize the feature vector afterwards or it is already normalized?

My image is of size 128x128. Should I use a win_Size of 128x128 as well? And how to select values for the other parameters as well: HOGDescriptor(Size win_size=Size(128, 128), Size block_size=Size(16, 16), Size block_stride=Size(8, 8), Size cell_size=Size(8, 8), int nbins=9, double win_sigma=DEFAULT_WIN_SIGMA, double threshold_L2hys=0.2, bool gamma_correction=true, int nlevels=DEFAULT_NLEVELS)

I have seen the explanation at http://docs.opencv.org/2.4/modules/gp... but it is not clear enough on how to set the parameters properly.

Thanks a lot.

I have extracted the following features of hand: compactness, aspect ration and orientation and stored them in a .csv file. Now I want to train them using KNN. KNN takes a matrix as input, how should I combines these features into a single Mat and train them.

I did not understand by what you want to say by improving it.

12 different classes and each class has 10 instances.

Thanks. Which classification method is more appropriate for chain code histogram feature?

I have already extracted chain code of the contour, calculated the number of occurence of each code. Then calculated the frequency of each code. Now I have to plot it in a histogram, I do not know how to create a histogram for chain code the no. of bins should be 8 thats all I know. But how to create it in opencv 3.0 using c++?

int main()
{
int count0 = 0;
int count1 = 0;
int count2 = 0;
int count3 = 0;
int count4 = 0;
int count5 = 0;
int count6 = 0;
int count7 = 0;

double totalCount = 0;

Mat image = imread("rectangle.jpg", 0);

Canny(image, image, 100, 100 * 2, 3, false);

CvChain* chain;
CvMemStorage* storage = 0;
storage = cvCreateMemStorage();

cvFindContours(&IplImage(image), storage, (CvSeq**)(&chain), sizeof(*chain), CV_RETR_EXTERNAL, CV_CHAIN_CODE);

int total = chain->total;


for (; chain != NULL; chain = (CvChain*)chain->h_next)
{

    int numChain = 0;

    CvSeqReader reader;
    int i, total = chain->total;

    cvStartReadSeq((CvSeq*)chain, &reader, 0);


    for (i = 0; i < total; i++)
    {
        char code;
        CV_READ_SEQ_ELEM(code, reader);
        int Fchain = (int)code;

        if (Fchain == 0)
        {
            count0++;
        }

        else if (Fchain == 1)
        {
            count1++;
        }

        else if (Fchain == 2)
        {
            count2++;
        }

        else if (Fchain == 3)
        {
            count3++;
        }

        else if (Fchain == 4)
        {
            count4++;
        }

        else if (Fchain == 5)
        {
            count5++;
        }

        else if (Fchain == 6)
        {
            count6++;
        }

        else if (Fchain == 7)
        {
            count7++;
        }

        totalCount++;
    }
}

cout << "0: " << count0 << endl;
cout << "1: " << count1 << endl;
cout << "2: " << count2 << endl;
cout << "3: " << count3 << endl;
cout << "4: " << count4 << endl;
cout << "5: " << count5 << endl;
cout << "6: " << count6 << endl;
cout << "7: " << count7 << endl;

cout << "Total: " << totalCount << endl;

double prob0 = (count0 / totalCount);
double prob1 = (count1 / totalCount);
double prob2 = (count2 / totalCount);
double prob3 = (count3 / totalCount);
double prob4 = (count4 / totalCount);
double prob5 = (count5 / totalCount);
double prob6 = (count6 / totalCount);
double prob7 = (count7 / totalCount);


cout << "Proababilty 0: " << prob0 << endl;
cout << "Proababilty 1: " << prob1 << endl;
cout << "Proababilty 2: " << prob2 << endl;
cout << "Proababilty 3: " << prob3 << endl;
cout << "Proababilty 4: " << prob4 << endl;
cout << "Proababilty 5: " << prob5 << endl;
cout << "Proababilty 6: " << prob6 << endl;
cout << "Proababilty 7: " << prob7 << endl;


waitKey(0);
return 0;

}