cv::Mat GaborFeatures::detectImageFeatures(const std::string& filename) {
cv::Mat img;
cv::Mat tmp;
cv::Mat histogram(1, orientationCount, CV_32FC1, 0.0f);
img = loadImage(filename);
cv::Mat maxVals(img.rows, img.cols, CV_32FC2, cv::Scalar(0, 0));
int i = 0;
for (float orientation = 0; orientation < M_PI; orientation += M_PI / orientationCount, i++) {
energyGabor(img, tmp, orientation);
// histogram.at<cv::Scalar>(0, column) = cv::sum(tmp);
for (int y = 0; y < img.rows; y++) {
for (int x = 0; x < img.cols; x++) {
if (tmp.at<float>(y, x) > maxVals.at<cv::Vec2f>(y, x)[0]) {
maxVals.at<cv::Vec2f>(y, x)[0] = tmp.at<float>(y, x);
maxVals.at<cv::Vec2f>(y, x)[1] = i;
}
}
}
}
for (int y = 0; y < img.rows; y++) {
for (int x = 0; x < img.cols; x++) {
histogram.at<float>(0, (int) maxVals.at<cv::Vec2f>(y, x)[1]) += 1;
}
}
return histogram;
}
void dataSummary(const char dataName[], int *nSamples, int *nAttributes)
{
int targetIndex;
Array<int> attNValues(1), hasMissingValues(1), isDiscrete(1);
Array<double> maxVals(1), minVals(1);
readDataInfo(dataName, nSamples, nAttributes, attNValues, hasMissingValues,
isDiscrete, maxVals, minVals, &targetIndex);
*nAttributes -= 1; // exclude target attribute
return;
}
cv::Mat GaborFeatures::detectImageFeatures(unsigned int imgNumber, unsigned int imgLabel) {
cv::Mat img;
cv::Mat tmp;
cv::Mat histogram(1, orientationCount, CV_32FC1, 0.0f);
std::stringstream filename;
if(imgLabel >= imageFilePaths.size()) {
std::cerr << "Label out of range!" << std::endl;
exit(EXIT_FAILURE);
}
filename << imageFilePaths[imgLabel] << "image_" << std::setw(4) << std::setfill('0') << imgNumber << ".jpg";
img = loadImage(filename.str());
cv::Mat maxVals(img.rows, img.cols, CV_32FC2, cv::Scalar(0, 0));
int i = 0;
for (float orientation = 0; orientation < M_PI; orientation += M_PI / orientationCount, i++) {
energyGabor(img, tmp, orientation);
// histogram.at<cv::Scalar>(0, column) = cv::sum(tmp);
for (int y = 0; y < img.rows; y++) {
for (int x = 0; x < img.cols; x++) {
if (tmp.at<float>(y, x) > maxVals.at<cv::Vec2f>(y, x)[0]) {
maxVals.at<cv::Vec2f>(y, x)[0] = tmp.at<float>(y, x);
maxVals.at<cv::Vec2f>(y, x)[1] = i;
}
}
}
}
for (int y = 0; y < img.rows; y++) {
for (int x = 0; x < img.cols; x++) {
histogram.at<float>(0, (int) maxVals.at<cv::Vec2f>(y, x)[1]) += 1;
}
}
return histogram;
}
// discardMissing means any data with 1 or more missing attributes will be
// discarded. Note that this may make the dataset returned inconsistent
// with what dataSummary() returns.
Data* new_readData(const char dataName[], int discardMissing)
{
Data *data;
int nSamples, nAttributes, targetIndex;
Array<int> attNValues(1), hasMissingValues(1), isDiscrete(1);
Array<double> maxVals(1), minVals(1);
readDataInfo(dataName, &nSamples, &nAttributes, attNValues, hasMissingValues,
isDiscrete, maxVals, minVals, &targetIndex);
// nAttributes counts target as an attribute as well.
// for class Data, we'll need to subtract 1 back off for data->nAttributes;
// allocate space for and set up internals of data
data = new Data(nSamples, nAttributes-1);
// nAttributes-1 because class Data does not count
// target as an attribute, whereas our file format does.
assert(data != NULL);
int dex=0;
for (int att=0; att < nAttributes; att++)
{
if (att == targetIndex)
{
data->targetIsContinuous = !isDiscrete[att];
assert(!hasMissingValues[att]); // missing values not allowed in target
// printf("[%d]", hasMissingValues[att]);
data->targetNValues = attNValues[att];
}
else
{
data->attIsContinuous[dex] = !isDiscrete[att];
data->hasMissingValues[dex] = hasMissingValues[att];
// printf("(%d)", hasMissingValues[att]);
data->attNValues[dex] = attNValues[att];
dex++;
}
}
assert(dex == nAttributes-1);
// start reading the data
char datFilename[4096];
strcpy(datFilename, DATADIR); strcat(datFilename, dataName);
strcat(datFilename, "/"); strcat(datFilename, dataName);
strcat(datFilename, ".dat"); // ".../data/foo/foo.dat"
FILE *fp = safe_fopen(datFilename, "rt");
int sampleDex=0, nHasMissingAtt=0;
for (int ctr=0; ctr < nSamples; ctr++)
{
int hasMissingAtt = readSample(fp, data, nAttributes, targetIndex, hasMissingValues, sampleDex);
if (discardMissing && hasMissingAtt)
nHasMissingAtt++;
else
sampleDex++;
}
fclose(fp);
// If some samples were discarded, then shrink the datastructures to reflect
// that we didn't actually get nSamples examples read in.
if (!discardMissing)
assert(nHasMissingAtt==0 && sampleDex==nSamples);
if (discardMissing)
{
data->nSamples = nSamples - nHasMissingAtt;
if (data->nSamples == 0)
error("new_readData: After discarding missing, have no data left!");
for (int dex=0; dex < data->nAttributes; dex++)
data->hasMissingValues[dex] = 0;
Matrix temp(data->nSamples, data->nAttributes);
for (int i=0; i < data->nSamples; i++) for (int j=0; j < data->nAttributes; j++)
temp[i][j] = data->dataSample[i][j];
data->dataSample.resize(data->nSamples, data->nAttributes);
data->dataSample = temp;
}
data->shuffleSample();
assert(data->nAttributes == nAttributes-1);
return data;
}
