FHOG.cpp

/*
C++ Felzenszwalb HOG extractor
 
This repository is meant to provide an easy-to-use implementation of the Felzenszwalb HOG features extractor.
This approach followed the one presented by Felzenszwalb, Pedro F., et al. "Object detection with discriminatively trained part-based models." Pattern Analysis and Machine Intelligence, IEEE Transactions on 32.9 (2010): 1627-1645. 
The OpenCV library have only the original HOG, proposed by Dalal and Triggs. However, the Latent SVM OpenCV implementation have its own FHOG extractor. This code allows you to use it without having do deal with Latent SVM objects.

To run the code you need OpenCV library.

Author: Joao Faro
Contacts: joaopfaro@gmail.com
*/

#include "FHOG.hpp"

using namespace std;
using namespace cv;

HogFeature::HogFeature(){

	_cell_size = 4;
	_scale = 1;

}

HogFeature::HogFeature(uint cell_size, uint scale){

	_cell_size = cell_size;
	_scale = scale;

}

HogFeature::~HogFeature(){

	freeFeatureMapObject(&_map);
	
}

HogFeature* HogFeature::clone() const{

	return new HogFeature(*this);

}

Mat HogFeature::getFeature(Mat image){

    _tmpl_sz.width = _scale * image.cols;
   	_tmpl_sz.height = _scale * image.rows;

    // Round to cell size and also make it even
    _tmpl_sz.width = ( ( (int)(_tmpl_sz.width / (2 * _cell_size)) ) * 2 * _cell_size ) + _cell_size*2;
    _tmpl_sz.height = ( ( (int)(_tmpl_sz.height / (2 * _cell_size)) ) * 2 * _cell_size ) + _cell_size*2;

    image.convertTo(image, CV_32F, 1 / 255.f); 
    if (image.cols != _tmpl_sz.width || image.rows != _tmpl_sz.height) {
        resize(image, image, _tmpl_sz);
    }   

    // Add extra cell filled with zeros around the image
    cv::Mat featurePaddingMat( _tmpl_sz.height+_cell_size*2, _tmpl_sz.width+_cell_size*2, CV_32FC3, cvScalar(0,0,0) );
    //image.copyTo(featurePaddingMat.rowRange(_cell_size, _cell_size+_tmpl_sz.height).colRange(_cell_size, _cell_size+_tmpl_sz.width));
    image.copyTo(featurePaddingMat);

    // HOG features
    IplImage zz = featurePaddingMat;
    getFeatureMaps(&zz, _cell_size, &_map);
    normalizeAndTruncate(_map, 0.2f);
    PCAFeatureMaps(_map);
    _featuresMap = Mat(Size(_map->numFeatures*_map->sizeX*_map->sizeY,1), CV_32F, _map->map);  // Procedure do deal with cv::Mat multichannel bug
    _featuresMap=_featuresMap.clone();
    freeFeatureMapObject(&_map);
    return _featuresMap;

}