double FacialFeatureRecognizer::compareFaces(Mat& face1, Mat& face2)
{
Mat eigenVectors = recognizer->getMat("eigenvectors");
Mat mean = recognizer->getMat("mean");
Mat proj = subspaceProject(eigenVectors, mean, face1.reshape(1,1));
Mat proj2 = subspaceProject(eigenVectors, mean, face2.reshape(1,1));
return norm(proj, proj2, NORM_L2);
}
void PCAKNN::train(const list<Interval> &intervals, const bool console_output) {
if (intervals.size() == 0) return;
projections.clear(); n = 0;
width = intervals.front().start->face.cols;
height = intervals.front().start->face.rows;
list<Interval>::const_iterator itr;
vector<Face>::const_iterator start, end;
//add the number of images from all intervals
for (itr = intervals.cbegin(); itr != intervals.cend(); itr++) {
n += itr->Length();
}
Mat pca_matrix(static_cast<int>(n), width*height, data_type);
int c = 0;
for (itr = intervals.cbegin(); itr != intervals.cend(); itr++) {
//for each image in the current interval
for (start = itr->start, end = itr->end; start != end; start++, ++c) {
if (console_output) printf("Preparing samples %d/%d\n", c + 1, n);
//insert current image into pca_matrix
Mat image_row = start->face.clone().reshape(1, 1);
Mat row_i = pca_matrix.row(c);
image_row.convertTo(row_i, data_type);//CV_64FC1 ?
Face f; f.name = start->name;
projections.push_back(f);//save the names for later
}
}
if (console_output) printf("TRAINING...\n");
//Perfrom principal component analysis on pca_matrix
PCA pca(pca_matrix, Mat(), CV_PCA_DATA_AS_ROW, pca_matrix.rows);
//extract mean/eigenvalues
mean = pca.mean.reshape(1, 1);
ev = pca.eigenvalues.clone();
transpose(pca.eigenvectors, w);
//project each face into subspace and save them with the name above for recognition
for (unsigned int i = 0; i<n; ++i) {
if (console_output) printf("Projecting %d/%d\n", i + 1, n);//project so subspace
projections[i].face = subspaceProject(w, mean, pca_matrix.row(i));
}
}
// Generate an approximately reconstructed face by back-projecting the eigenvectors & eigenvalues of the given (preprocessed) face.
Mat reconstructFace(const Ptr<FaceRecognizer> model, const Mat preprocessedFace)
{
// Since we can only reconstruct the face for some types of FaceRecognizer models (ie: Eigenfaces or Fisherfaces),
// we should surround the OpenCV calls by a try/catch block so we don't crash for other models.
try {
// Get some required data from the FaceRecognizer model.
Mat eigenvectors = model->get<Mat>("eigenvectors");
Mat averageFaceRow = model->get<Mat>("mean");
int faceHeight = preprocessedFace.rows;
// Project the input image onto the PCA subspace.
Mat projection = subspaceProject(eigenvectors, averageFaceRow, preprocessedFace.reshape(1,1));
//printMatInfo(projection, "projection");
// Generate the reconstructed face back from the PCA subspace.
Mat reconstructionRow = subspaceReconstruct(eigenvectors, averageFaceRow, projection);
//printMatInfo(reconstructionRow, "reconstructionRow");
// Convert the float row matrix to a regular 8-bit image. Note that we
// shouldn't use "getImageFrom1DFloatMat()" because we don't want to normalize
// the data since it is already at the perfect scale.
// Make it a rectangular shaped image instead of a single row.
Mat reconstructionMat = reconstructionRow.reshape(1, faceHeight);
// Convert the floating-point pixels to regular 8-bit uchar pixels.
Mat reconstructedFace = Mat(reconstructionMat.size(), CV_8U);
reconstructionMat.convertTo(reconstructedFace, CV_8U, 1, 0);
//printMatInfo(reconstructedFace, "reconstructedFace");
return reconstructedFace;
} catch (cv::Exception e) {
//cout << "WARNING: Missing FaceRecognizer properties." << endl;
return Mat();
}
}
//k should be uneven to break ties
string PCAKNN::recognize(const Mat &face, const unsigned int k, const bool use_distance_weighting, const bool console_output) const {
string name = "N/A";
if (n <= k) return name;
//project target face to subspace
Mat target = subspaceProject(w, mean, face.reshape(1, 1));
vector<unsigned int> classes(k, 0);
vector<double> distances(k, DBL_MAX);
double dist = DBL_MAX;
//find k nearest neighbours
for (unsigned int i = 0; i < projections.size(); ++i) {
dist = norm(projections[i].face, target, NORM_L2);//norml2
for (unsigned int j = 0; j < k; ++j) {
if (dist < distances[j]) {
//discard the worst match and shift remaining down
for (int l = k - 1; l > j; --l) {
distances[l] = distances[l - 1];
classes[l] = classes[l - 1];
}
classes[j] = i; //set new best match
distances[j] = dist;
break;
}
}
}
map<string, Weighting> neighbours;
//count occurence of classes
for (unsigned int i = 0; i<k; ++i) {
Weighting &weight = neighbours[projections[classes[i]].name];
weight.count++;
weight.total_dist += distances[i];
}
//evaluate voting
if (use_distance_weighting) {
double min_weight = DBL_MAX;
for (map<string, Weighting>::iterator itr = neighbours.begin(); itr != neighbours.end(); ++itr) {
double weight = itr->second.total_dist / (double)itr->second.count;
//concider average weight instead of number of votes
if (weight < min_weight) {
min_weight = weight;
name = itr->first;
}
if (console_output)printf("%s %f\n", itr->first.c_str(), weight);
}
}
else {
unsigned int max_count = 0;
//choose the class with the most votes
for (map<string, Weighting>::iterator itr = neighbours.begin(); itr != neighbours.end(); ++itr) {
if (itr->second.count > max_count) {
max_count = itr->second.count;
name = itr->first;
}
if (console_output)printf("%s %d\n", itr->first.c_str(), itr->second.count);
}
}
return name;
}
