EmguCv使用哪种策略进行SVM分类？

Question

Currently I am doing my academic Project (Social Media Website). My intention is when a user post an image the system should recognize the face and label it with her name.

As prior work I created 5 users. For training Images, when a user set her profile photo the system will detect it using EmguCv (DetectHaarCascade) and save the image as a bitmap in a folder. The label of that face will be the User Id of the User, and the label is saved in a text file within the folder. As training images I uploaded and labeled 10 images for each user.

The next part is when a user post a photo. The system should recognize the face and label it with her name. I am using SVM for recognition and classification.

My detection code:

    //Face Detection
    MCvAvgComp[][] facesDetected = gray.DetectHaarCascade(
                      haar,
                      1.2,
                      4,
                      HAAR_DETECTION_TYPE.DO_CANNY_PRUNING,
                      new Size(20, 20));
    foreach (MCvAvgComp f in facesDetected[0])
    {
     result = currentFrame.Copy(f.rect).Convert<Gray, byte>().Resize(100, 100, Emgu.CV.CvEnum.INTER.CV_INTER_CUBIC);
     // at the beginning i am added all the existing images to an List<Image<Gray, byte>> trainingImages and labels to List<string> labels 
     trainingImages.Add(result );
     labels.Add(Session["UserId"].ToString())
     File.WriteAllText((Server.MapPath("~/TrainedFaces/TrainedLabels.txt")), trainingImages.ToArray().Length.ToString() + "%");

       for (int i = 1; i < trainingImages.ToArray().Length + 1; i++)
       {
        //saving the trained images and labells
        trainingImages.ToArray()[i - 1].Save(Server.MapPath("~/TrainedFaces/") + "face" + i + ".bmp");
        File.AppendAllText(Server.MapPath("~/TrainedFaces/TrainedLabels.txt"), labels.ToArray()[i - 1] + "%");
       }
   }

My SVM training code:

/*
1. Loaded all the images to trainingImages
2.Loaded all the labels to labels
*/

// Converting My labesl and images to Matrix for preparing training data and training label    

     Matrix<float> TrainindData = new Matrix<float>(trainingImages.Count, 100 * 100);
        int ii = 0;

        foreach (Image<Gray, float> img in trainingImages)
        {
            int jj = 0;
            Matrix<float> Imagemtrx = new Matrix<float>(img.Width, img.Height);
            img.CopyTo(Imagemtrx);
            for (int k = 0; k < Imagemtrx.Rows; k++)
            {
                for (int j = 0; j < Imagemtrx.Cols; j++)
                {
                    TrainindData.Data[ii, jj] = Imagemtrx[k, j];
                    jj++;
                }
            }
            ii++;
        }
        Matrix<float> TrainedLabels = new Matrix<float>(labels.Count, 1);
        int kk = 0;
        foreach (int lab in labels)
        {
            TrainedLabels[kk, 0] = lab;
            kk++;
        }

     SVM model = new SVM();

     SVMParams p = new SVMParams();
        p.KernelType = Emgu.CV.ML.MlEnum.SVM_KERNEL_TYPE.LINEAR;
        //p.SVMType = Emgu.CV.ML.MlEnum.SVM_TYPE.ONE_CLASS;
        p.SVMType = Emgu.CV.ML.MlEnum.SVM_TYPE.C_SVC;
        p.C = 1;
        p.TermCrit = new MCvTermCriteria(100, 0.00001);

        bool trained = model.Train(TrainindData, TrainedLabels, null, null, p);

My recognition and classification code:

MCvAvgComp[][] faces=grayFrame.DetectHaarCascade(haar,1.2,10,HAAR_DETECTION_TYPE.DO_CANNY_PRUNING,new Size(20, 20));
foreach (MCvAvgComp face in faces[0])
        {
            InputFrame.Draw(face.rect, new Bgr(Color.Red), 1);
            t = t + 1;
            Image<Gray, float> result = InputFrame.Copy(face.rect).Convert<Gray, float>().Resize(100, 100, Emgu.CV.CvEnum.INTER.CV_INTER_CUBIC);
            //result._EqualizeHist();
            Matrix<float> TestImageMatix = new Matrix<float>(result.Width, result.Height);
            result.CopyTo(TestImageMatix);
            Matrix<float> TestData = new Matrix<float>(1, result.Width * result.Height);
            int z = 0;
            for (int k = 0; k < TestImageMatix.Rows; k++)
            {
                for (int j = 0; j < TestImageMatix.Cols; j++)
                {
                    TestData.Data[0, z] = TestImageMatix[k, j];
                    z++;
                }
            }
            //Here I will Get the UserId as class label. I can find name from database using this Id
            float result1 = model.Predict(TestData);

Now the problem is that when I upload an image that belongs to any of existing class, it will correctly recognize and identify the person. But when I post a different photo (of a person that is not there in social media), then it's assigned to one of the existing class label.

My questions are:

1. I Want to identify only the right person. Remaining can be labeled as Unknown or something else (I dont Know whether any other method needed or not)

2. I read about One Vs One and One Vs All Strategies. Which one is used in my code?

3. If no one is used then how to implement them?.

4. Emgu CV already includes SVM. Which type is used in It?

Answer 1

I read about One Vs One and One Vs All Strategies. Which one is used in my code? If no one is used then how to implement them? Emgu CV already includes SVM. Which type is used in it?

You are using One vs All strategy, which is implemented in OpenCV using n*(n-1)/2 One Vs One classifiers ( n is the number of labels).

I Want to Identify only the right person. Remaining can be labeled as Unknown or something else( I dont Know whether any other method needed or not)

You can built n*(n-1)/2 One Vs One classifiers, and then get the raw response from predict (works only with 2-class problems). The majority of responses identify the output class. If the maximum response for the classified class is under a threshold, you can say that none of the class if correctly recognized.

Or you can use n One class classifiers, and again check if the response for each is under a given threshold.