//Bayesian Classifier
void bayes ( Mat & trainingData , Mat & trainingClasses , Mat & testData , Mat
& testClasses ) {
CvNormalBayesClassifier bayes ( trainingData , trainingClasses ) ;
Mat predicted ( testClasses.rows , 1 , CV_32F ) ;
for ( int i = 0; i < testData.rows ; i ++) {
const Mat sample = testData.row ( i ) ;
predicted.at < float > (i , 0) = bayes.predict ( sample ) ;
}
cout << " Accuracy_ { BAYES } = " << evaluate ( predicted , testClasses ) << endl ;
plot_binary ( testData , predicted , " Predictions Bayes " ) ;
}
void Model::Predict_nbayes( const SampleSet& samples, SampleSet& outError )
{
int true_resp = 0;
CvNormalBayesClassifier* model = (CvNormalBayesClassifier*)m_pModel;
cv::Mat result;
model->predict(samples.Samples(),&result);
for (int i = 0; i < samples.N(); i++)
{
if (result.at<float>(i) != samples.GetLabelAt(i))
{
outError.Add(samples.GetSampleAt(i), samples.GetLabelAt(i));
}
else
{
true_resp++;
}
}
printf("%d %d",samples.N(), true_resp);
}
int _tmain(int argc, _TCHAR* argv[])
{
// General training of a naive bayes model - trafic sign model training in this specific case
// A model is trained by supplying positive and negative examples and letting the classifier decide how to seperate them
// For this training, a feature vector of each sample and an output needs to be supplied to the algoritm
// Read in file, in order to be able to segment out the wrong ones
if( argc == 1 ){
printf( "Usage: Application trains a naive bayes classifier on positive and negative samples.\n"
"For the moment parameters and feature construction are specified in the code.\n"
"--> naive_bayes_model_training.exe\n");
return 0;
}
// Parameters - comment out
string main_folder = "C:/data/traffic_signs/";
string positive_folder = main_folder + "positives/";
string negative_folder = main_folder + "negatives/";
vector<Mat> input_elements, input_elements_cmy, features;
CvNormalBayesClassifier classifier;
// Read in the positive and negative training samples and put them into a specific vector
// C/C++ doesn't provide a standard file reading system, this is WINDOWS based and will not work under Linux/MacOsx/Android/... !
WIN32_FIND_DATA ffd;
TCHAR szDir_positives[MAX_PATH], szDir_negatives[MAX_PATH];
vector<string> filenames_positives, filenames_negatives;
HANDLE hFind = INVALID_HANDLE_VALUE;
StringCchCopy(szDir_positives, MAX_PATH, positive_folder.c_str());
StringCchCat(szDir_positives, MAX_PATH, TEXT("*"));
StringCchCopy(szDir_negatives, MAX_PATH, negative_folder.c_str());
StringCchCat(szDir_negatives, MAX_PATH, TEXT("*"));
// Process data from the positives folder, at the end remove the current and parent folder references in the listing
hFind = FindFirstFile(szDir_positives, &ffd);
do
{
filenames_positives.push_back (ffd.cFileName);
}
while (FindNextFile(hFind, &ffd) != 0);
filenames_positives.erase(filenames_positives.begin(),filenames_positives.begin()+2);
FindClose( hFind );
// Process data from the negatives folder, at the end remove the current and parent folder references in the listing
hFind = FindFirstFile(szDir_negatives, &ffd);
do
{
filenames_negatives.push_back (ffd.cFileName);
}
while (FindNextFile(hFind, &ffd) != 0);
filenames_negatives.erase(filenames_negatives.begin(),filenames_negatives.begin()+2);
FindClose( hFind );
// Based on the filenames, read in the data and put into the matrix vector
for(int i = 0; i < filenames_positives.size(); i++){
Mat temp = imread(filenames_positives[i]);
input_elements.push_back(temp);
}
for(int i = 0; i < filenames_negatives.size(); i++){
Mat temp = imread(filenames_negatives[i]);
input_elements.push_back(temp);
}
// Switch inputs from bgr to cmy color space
for(int i = 0; i < 10; i++){
Mat temp = bgr_to_cmy(input_elements[i]);
input_elements_cmy.push_back(temp);
}
// Pick the yellow channel of each sample, create a feature representation of it and return to the naïve bayes classifier
for(int i = 0; i < input_elements.size(); i++){
Mat temp_feature = create_feature_cmy(input_elements_cmy[i]);
features.push_back(temp_feature);
}
// Create feature matrix based on retrieved vector with the single features
Mat input_features = convert_input_vector(features);
// Create according feature label matrix
Mat label_features(10,1,CV_8UC1);
for(int i = 0; i < filenames_positives.size(); i++){
label_features.push_back(1);
}
for(int i = 0; i < filenames_negatives.size(); i++){
label_features.push_back(0);
}
// Train the normal bayes classifier model and store in into an xml file so that it can be reused
// Training data and labels need to be of the CV_32FC1 format!
input_features.convertTo(input_features, CV_32FC1);
label_features.convertTo(label_features, CV_32FC1);
classifier.train(input_features, label_features);
// Save the normal bayes classifer model
string output_location = main_folder + "naive_bayes_traffic_signs.xml";
classifier.save(main_folder.c_str());
return 0;
}
void Model::Train_nbayes( const SampleSet& samples )
{
CvNormalBayesClassifier* model = (CvNormalBayesClassifier*)m_pModel;
//void* para = (void*)m_trainPara;
model->train(samples.Samples(), samples.Labels());
}
void Naivebayesopencv::test()
{
//testPCA();
//number of taining samples to be used
int numImages = 1000;
CvMat *trainingVectors = 0;
CvMat *trainingLabels = 0;
extractTrainingData(numImages, trainingVectors, trainingLabels);
//train the data with Naivebayes
// Perform a PCA:
PCA pca(Mat(trainingVectors), Mat(), CV_PCA_DATA_AS_ROW, 300);
Mat newTraining = pca.project(Mat(trainingVectors));
CvNormalBayesClassifier Naivebayes;
Naivebayes.train(newTraining, trainingLabels, Mat(), Mat(), false);
/*Naivebayes.train(trainingVectors, trainingLabels, Mat(), Mat(), false);*/
cvReleaseMat(&trainingVectors);
cvReleaseMat(&trainingLabels);
//Recognition: Using Naivebayes
//test number to be used
numImages = 1000;//for tesing number
CvMat *testVectors = 0;
CvMat *actualLabels = 0;
extractTestingData(numImages, testVectors, actualLabels);
CvMat *testLabels = cvCreateMat(numImages, 1, CV_32FC1);
PCA pca2(Mat(testVectors), Mat(), CV_PCA_DATA_AS_ROW, 300);
Mat newTesignVector = pca2.project(Mat(testVectors));
CvMat cvMatnewTestingvector = newTesignVector;
Naivebayes.predict(&cvMatnewTestingvector, testLabels);
//Naivebayes.predict(testVectors, testLabels);
//Get the error rate and print the result out
int totalCorrect = 0;
for (int i = 0; i < numImages; i++){
if (testLabels->data.fl[i] == actualLabels->data.fl[i])
{
totalCorrect++;
}
else{
printf("\n Error: image id=%d number %f was mistaken as %f", i, actualLabels->data.fl[i], testLabels->data.fl[i]);
}
}
printf("\nError Rate: %.1f%%",
(double)100- (double)totalCorrect * 100 / (double)numImages);
cvReleaseMat(&testVectors);
cvReleaseMat(&testLabels);
cvReleaseMat(&actualLabels);
return;
}
int predict(const CvNormalBayesClassifier& _bayes, const cv::Mat& _mat) {
return (int) _bayes.predict(_mat);
}
void train(cMatRef _training_data, cMatRef _label ) {
CvNormalBayesClassifier bayes;
bayes.train(_training_data, _label);
bayes.save("model.yml");
}
