我想通过在C ++和Visual Studio 2013中使用opencv 2.4.11库来学习svm实现。我的代码：
#include "opencv2\highgui\highgui.hpp"
#include <iostream>
#include "opencv2\core\core.hpp"
#include "opencv2\imgproc\imgproc.hpp"
#include "opencv2/ml/ml.hpp"
using namespace cv; // OpenCV API is in the C++ "cv" namespace
#include <stdio.h>
/******************************************************************************/
// use SVM "grid search" for kernel parameters
#define USE_OPENCV_GRID_SEARCH_AUTOTRAIN 1 // set to 0 to set SVM parameters manually
/******************************************************************************/
#define NUMBER_OF_TRAINING_SAMPLES 6238
#define ATTRIBUTES_PER_SAMPLE 617
#define NUMBER_OF_TESTING_SAMPLES 1559
#define NUMBER_OF_CLASSES 26
// N.B. classes are spoken alphabetric letters A-Z labelled 1 -> 26
/******************************************************************************/
// loads the sample database from file (which is a CSV text file)
int read_data_from_csv(const char* filename, Mat data, Mat classes, int n_samples)
{
float tmp;
// if we can't read the input file then return 0
FILE* f = fopen(filename, "r");
//printf("asdadad\n");
if (!f)
{
printf("ERROR: cannot read file %s\n", filename);
return 0; // all not OK
}
// for each sample in the file
for (int line = 0; line < n_samples; line++)
{
// for each attribute on the line in the file
for (int attribute = 0; attribute < (ATTRIBUTES_PER_SAMPLE + 1); attribute++)
{
if (attribute < ATTRIBUTES_PER_SAMPLE)
{
// first 617 elements (0-616) in each line are the attributes
fscanf(f, "%f,", &tmp);
data.at<float>(line, attribute) = tmp;
}
else if (attribute == ATTRIBUTES_PER_SAMPLE)
{
// attribute 617 is the class label {1 ... 26} == {A-Z}
fscanf(f, "%f,", &tmp);
classes.at<float>(line, 0) = tmp;
}
}
}
fclose(f);
return 1; // all OK
}
/******************************************************************************/
int main(int argc, char** argv)
{
// lets just check the version first
printf("OpenCV version %s (%d.%d.%d)\n",
CV_VERSION,
CV_MAJOR_VERSION, CV_MINOR_VERSION, CV_SUBMINOR_VERSION);
// define training data storage matrices (one for attribute examples, one
// for classifications)
Mat training_data = Mat(NUMBER_OF_TRAINING_SAMPLES, ATTRIBUTES_PER_SAMPLE, CV_32FC1);
Mat training_classifications = Mat(NUMBER_OF_TRAINING_SAMPLES, 1, CV_32FC1);
//define testing data storage matrices
Mat testing_data = Mat(NUMBER_OF_TESTING_SAMPLES, ATTRIBUTES_PER_SAMPLE, CV_32FC1);
Mat testing_classifications = Mat(NUMBER_OF_TESTING_SAMPLES, 1, CV_32FC1);
// load training and testing data sets
if (read_data_from_csv("Project2.csv", training_data, training_classifications, NUMBER_OF_TRAINING_SAMPLES) &&
read_data_from_csv("Project2.csv", testing_data, testing_classifications, NUMBER_OF_TESTING_SAMPLES))
{
// define the parameters for training the SVM (kernel + SVMtype type used for auto-training,
// other parameters for manual only)
CvSVMParams params = CvSVMParams(
CvSVM::C_SVC, // Type of SVM, here N classes (see manual)
CvSVM::LINEAR, // kernel type (see manual)
0.0, // kernel parameter (degree) for poly kernel only
0.0, // kernel parameter (gamma) for poly/rbf kernel only
0.0, // kernel parameter (coef0) for poly/sigmoid kernel only
10, // SVM optimization parameter C
0, // SVM optimization parameter nu (not used for N classe SVM)
0, // SVM optimization parameter p (not used for N classe SVM)
NULL, // class wieghts (or priors)
// Optional weights, assigned to particular classes.
// They are multiplied by C and thus affect the misclassification
// penalty for different classes. The larger weight, the larger penalty
// on misclassification of data from the corresponding class.
// termination criteria for learning algorithm
cvTermCriteria(CV_TERMCRIT_ITER + CV_TERMCRIT_EPS, 1000, 0.000001)
);
// train SVM classifier (using training data)
printf("\nUsing training database: %s\n\n", argv[1]);
CvSVM* svm = new CvSVM;
printf("\nTraining the SVM (in progress) ..... ");
fflush(NULL);
#if (USE_OPENCV_GRID_SEARCH_AUTOTRAIN)
printf("(SVM 'grid search' => may take some time!)");
fflush(NULL);
// train using auto training parameter grid search if it is available
// (i.e. OpenCV 2.x) with 10 fold cross valdiation
// N.B. this does not search kernel choice
svm->train_auto(training_data, training_classifications, Mat(), Mat(), params, 10);
params = svm->get_params();
printf("\nUsing optimal parameters degree %f, gamma %f, ceof0 %f\n\t C %f, nu %f, p %f\n Training ..",
params.degree, params.gamma, params.coef0, params.C, params.nu, params.p);
#else
// otherwise use regular training and use parameters manually specified above
svm->train(training_data, training_classifications, Mat(), Mat(), params);
#endif
printf(".... Done\n");
// get the number of support vectors used to define the SVM decision boundary
printf("Number of support vectors for trained SVM = %i\n", svm->get_support_vector_count());
// perform classifier testing and report results
Mat test_sample;
int correct_class = 0;
int wrong_class = 0;
int false_positives[NUMBER_OF_CLASSES];
char class_labels[NUMBER_OF_CLASSES];
float result;
// zero the false positive counters in a simple loop
for (int i = 0; i < NUMBER_OF_CLASSES; i++)
{
false_positives[i] = 0;
class_labels[i] = (char)65 + i; // ASCII 65 = A
}
printf("\nUsing testing database: %s\n\n", argv[2]);
for (int tsample = 0; tsample < NUMBER_OF_TESTING_SAMPLES; tsample++)
{
// extract a row from the testing matrix
test_sample = testing_data.row(tsample);
// run SVM classifier
result = svm->predict(test_sample);
// printf("Testing Sample %i -> class result (character %c)\n", tsample, class_labels[((int) result) - 1]);
// if the prediction and the (true) testing classification are the same
// (N.B. openCV uses a floating point decision tree implementation!)
if (fabs(result - testing_classifications.at<float>(tsample, 0))
>= FLT_EPSILON)
{
// if they differ more than floating point error => wrong class
wrong_class++;
false_positives[(int)(testing_classifications.at<float>(tsample, 0) - 1)]++;
}
else
{
// otherwise correct
correct_class++;
}
}
printf("\nResults on the testing database: %s\n"
"\tCorrect classification: %d (%g%%)\n"
"\tWrong classifications: %d (%g%%)\n",
argv[2],
correct_class, (double)correct_class * 100 / NUMBER_OF_TESTING_SAMPLES,
wrong_class, (double)wrong_class * 100 / NUMBER_OF_TESTING_SAMPLES);
for (unsigned char i = 0; i < NUMBER_OF_CLASSES; i++)
{
printf("\tClass (character %c) false postives %d (%g%%)\n", class_labels[(int)i],
false_positives[(int)i],
(double)false_positives[i] * 100 / NUMBER_OF_TESTING_SAMPLES);
}
// all matrix memory free by destructors
// all OK : main returns 0
return 0;
}
// not OK : main returns -1
return -1;
}
跑完后：
(SVM 'grid search' => may take some time!)
Opencv Error: Bad argument (response #0 is not integral) in
cvPreprocessCategoricalResponses,
file C:\builds\2_4_PackSlave-win64-vc12-shared\opencv\modules\ml\src\inner_function.cpp, line 715
我调试了那段代码。调试器在此行停止：
line：svm-＆gt; train_auto（training_data，training_classifications，Mat（）， Mat（），params，10）;