I have written following program..
#include "stdafx.h"
#include <conio.h>
#include <iostream>
#include <iomanip>
#include <opencv2/opencv.hpp>
#include <opencv2\core\core.hpp>
#include <opencv2\imgproc\imgproc.hpp>
using namespace cv;
using namespace ml;
using namespace std;
//HOG vector catenation per patch done!!
//
vector <Mat> HOGvector;
Mat histogram(Mat angle,Mat mag)
{
if(!angle.empty() && !mag.empty())
{
Mat hist=Mat::zeros(1,9,CV_32FC1);
for(int r=0;r<angle.rows;r++)
{
for(int c=0;c<angle.cols;c++)
{
float a=(angle.at<float>(r,c));
int bin=int( (int)a % 180)/20;
hist.at<float>(0,bin)+=mag.at<float>(r,c);
}
}
normalize(hist,hist);
return hist;
}
else
{
return Mat::zeros(1,9,CV_32FC1);
}
}
void main()
{
puts("lol");
Mat h1,h2,h3;
CascadeClassifier face("haarcascade_frontalface_alt.xml");
if(face.empty())
{
puts("!!CASCADE LOADING FAILURE...");
exit(0);
}
else{}
Mat image_=imread("lena.jpg");
Mat tmp(image_);
imshow("tmp",tmp);
Mat image_grey;
cvtColor(tmp,image_grey,COLOR_BGR2GRAY);
equalizeHist(image_grey,image_grey);
vector<Rect>faces;
face.detectMultiScale(image_grey,faces,1.1,4,0|CASCADE_SCALE_IMAGE,Size(100,100));
int s=faces.size();
vector<Mat>Ptr;
printf("\nsize...:%d",s);
Mat ROI;
if(s>0)
{
for(int size=0;size<s;size++)
{
rectangle(tmp,Point(faces[size].x,faces[size].y),Point(faces[size].x+faces[size].width,faces[size].y+faces[size].height),Scalar(100,faces[size].height,faces[size].width),2);
ROI=Mat(tmp,faces[size]);
}
}
else
{
}
imshow("ROI",ROI);
printf("ROI DIMENSIONS: %d %d",ROI.rows,ROI.cols);
Mat image(ROI);
imshow("out",image);
image.convertTo(image,CV_32F,1/255.0);
Mat dx,dy; //image axial derivatives..
Sobel(image,dx,CV_32F,1,0,1);
Sobel(image,dy,CV_32F,0,1,1);
imshow("dx",dx);
imshow("dy",dy);
Mat mag,angle;
cartToPolar(dx,dy,mag,angle,true); //polar coordinate system with 'degree' as measurement of angle(not in rad.");
imshow("magnitude",mag);
imshow("angle",angle);
//now time to calculate 8x8 patch of transformed image
Mat patch;
int count=0;
int a=0;
int patchsize=8;
int iter=(int)image.rows/patchsize; //here iter defines no of iteration of patch over row or coloum..to estimate total iteration we need to square oter..
printf("iteration...%d",iter*iter);
printf("init Hog vector of size:%d",iter*iter);
HOGvector=vector<Mat>(iter*iter);
int r=0,c=0;
if(mag.rows<patchsize && mag.cols<patchsize)
{
puts("PATCH SIZE..LOL!@@#%");
_getch(); exit(0);
}
for(int i=1,r=0;i<=iter;i++,r+=8)
{
for(int j=1,c=0;j<=iter;j++,c+=8)
{
Mat Patch_angle=Mat(angle,Rect(r,c,8,8));
Mat Patch_mag=Mat(mag,Rect(r,c,8,8));
HOGvector[count]=histogram(Patch_angle,Patch_mag);
count+=1;
}
}
cvtColor(image,image,CV_BGR2GRAY);
printf("\ntotalfeatures:%d",count*9);
printf("\nHOG vector size:%d\n",HOGvector.size());
printf("Hog discriptor size:%d",HOGvector.size());
imshow("magniture(POLAR)",mag);
printf("\npixle lost during floating point adjustment..\n%d row %d cols",(image.rows-iter*patchsize),(image.cols-iter*patchsize));
imshow("single channel(Reduced version)",image);
mag.deallocate(); //flush all data..
angle.deallocate();
Mat HOGfeature(count,9,CV_32FC1);
printf("hog rows:%d hog cols:%d",HOGfeature.rows,HOGfeature.cols);
for(int r=0;r<HOGfeature.rows;r++)
{
for(int c=0;c<HOGfeature.cols;c++)
{
HOGfeature.at<float>(r,c)=HOGvector[r].at<float>(0,c);
HOGvector[r].deallocate();
}
}
//catenated HOG feature obtained..
//time to apply PCA algorithm over feature..
printf("\ninit PCA object..");
PCA pca(HOGfeature,Mat(),PCA::DATA_AS_ROW,1);
std::cout<<"eigen values:\n"<<pca.eigenvalues;
std::cout<<"\neigen vectors:\n"<<pca.eigenvectors;
std::cout<<"\nmean:\n"<<pca.mean;
printf("size:%d",pca.project(HOGfeature).size);
waitKey(0);
HOGfeature.deallocate();
_getch();
}
if you saw output of the program..(as I requested for single dimension out of all)I got one Eigen value and its corresponding Eigen vector..finally I actually got 11,025 dimension of catenated HOG vector of whole image(actually array of histogram )...then I fed this huge vector into a matrix (variable:HOGfeature) ..each row represent 9-bin histogram and whole matrix represent HOG vector of Whole image..Then I fed those vector to PCA element and got a 9-dimension Eigen vector....I just want to know I fed 11,025 vectors each of 9 dimension...and I obtained single Eigen value and single Eigen vector corresponding to that then what does that mean???
