具有BGR训练样本的OpenCV SVM异常

Question

我遇到了OpenCV SVM的问题，特别是EmguCV的SVM，一个用于C＃的OpenCV包装器。

我正在尝试使用SVM分类器查找图像区域。为此，我将图像通道用作3D协调空间中的点。

The SVM example from emguCV显示了一个2维分类器，但由于3个图像通道，我需要一个3维分类器：BGR。

下面的代码不会抛出异常：

var trainData = new Matrix<float>(trainSampleCount, 2);

但是，此代码在方法TrainAuto中引发异常：

var trainData = new Matrix<float>(trainSampleCount, 3);

例外是：

样本量与培训时间不同

有人能告诉我这是什么问题吗？

完整代码：

public Image<Bgr, byte> Process(Image<Bgr, byte> image)
{
    const int kFold = 4;

    var svm = new SVM();

    var p = new SVMParams
    {
        KernelType = Emgu.CV.ML.MlEnum.SVM_KERNEL_TYPE.LINEAR,

        SVMType = Emgu.CV.ML.MlEnum.SVM_TYPE.C_SVC,

        C = 1,

        TermCrit = new MCvTermCriteria(100, 0.001)
    };

    RaVec3[] beach = {
                      new RaVec3(182, 240, 251),
                          new RaVec3(187, 210, 228),
                          new RaVec3(187, 210, 228),
                          new RaVec3(191, 215, 236)
                      };
    RaVec3[] montain = {
                      new RaVec3(44, 51, 52),
                          new RaVec3(43, 51, 52),
                          new RaVec3(84, 88, 69),
                          new RaVec3(36, 36, 26)
                      };
    RaVec3[] ocean = {
                      new RaVec3(149,148,101),
                          new RaVec3(245, 221, 195),
                          new RaVec3(149,147,101),
                          new RaVec3(148,145,92)
                      };
    RaVec3[] sky = {
                          new RaVec3(253,253,240),
                          new RaVec3(252,246,218),
                          new RaVec3(243, 253, 254),
                          new RaVec3(251,246,218)
                  };

    var trainSampleCount = 4 * 4;

    //Works with 2-dimmensional space, but throws an exceptions with 3-dimmensional space
    var trainData = new Matrix<float>(trainSampleCount, 3);
    var trainClasses = new Matrix<float>(trainSampleCount, 1);

        var count = 0;
        for (var i = 0; i < 4; i++ ) {
            trainData[count, 0] = (float)beach[i].X;
            trainData[count, 1] = (float)beach[i].Y;
            trainData[count, 2] = (float)beach[i].Z;
            trainClasses[count, 0] = 0;
            count++;
        }
        for (var i = 0; i < 4; i++) {
            trainData[count, 0] = (float)montain[i].X;
            trainData[count, 1] = (float)montain[i].Y;
            trainData[count, 2] = (float)montain[i].Z;
            trainClasses[count, 0] = 1;
            count++;
        }
        for (var i = 0; i < 4; i++) {
            trainData[count, 0] = (float)ocean[i].X;
            trainData[count, 1] = (float)ocean[i].Y;
            trainData[count, 2] = (float)ocean[i].Z;
            trainClasses[count, 0] = 2;
            count++;
        }
        for (var i = 0; i < 4; i++) {
            trainData[count, 0] = (float)sky[i].X;
            trainData[count, 1] = (float)sky[i].Y;
            trainData[count, 2] = (float)sky[i].Z;
            trainClasses[count, 0] = 3;
            count++;
        }

        //Here, the exception is thrown
        svm.TrainAuto(trainData, trainClasses, null, null, p.MCvSVMParams, kFold);

        for (var i = 0; i < image.Height; i++)
        {
            for (var j = 0; j < image.Width; j++)
            {
                var sample = new Matrix<float>(1, 2);
                sample.Data[0, 0] = j;
                sample.Data[0, 1] = i;

                var response = (int)svm.Predict(sample);

                image[i, j] =
                    response == 0 ? new Bgr(182, 240, 251) :
                    response == 1 ? new Bgr(44, 51, 52) :
                    response == 2 ? new Bgr(149, 148, 10) :
                    new Bgr(253, 253, 240);
            }
        }

        var c = svm.GetSupportVectorCount();
        for (var i = 0; i < c; i++)
        {
            var v = svm.GetSupportVector(i);
            var p1 = new PointF(v[0], v[1]);
            image.Draw(new CircleF(p1, 4), new Bgr(128, 128, 128), 2);
        }

        return image;
    }