Consider, I have the following matrix
0 1 2 3
4 5 6 7
8 9 10 11
12 13 14 15
I want to retrieve the values in even indices (both x and y indices are even) without using for loop.
0 2
8 10
I have big sized images (many of 5000*5000+ grayscale matrices). Using for loop doesn't seem to be the best way. I'd like to hear if there is better way than for loops.
I tried using the following mask, then do the operations but it is not efficient because I need to do 4*n^2 multiplication rather than n^2(Assume original image is 2n*2n)
1 0 1 0
0 0 0 0
1 0 1 0
0 0 0 0
Note that, I do multiple operations on the matrix. Any help is appreciated.
Thanks in advance,
You can remove the useless rows and columns, and work on a matrix with half the size of the original matrix.
You can do this easily with the resize
function, with nearest interpolation:
#include <opencv2/opencv.hpp>
#include <iostream>
using namespace cv;
using namespace std;
int main(int argc, char **argv)
{
Mat1b mat = (Mat1b(4,4) << 0, 1, 2, 3,
4, 5, 6, 7,
8, 9, 10, 11,
12, 13, 14, 15);
Mat1b res;
resize(mat, res, Size(0, 0), 0.5, 0.5, INTER_NEAREST);
cout << "Mat:" << endl << mat << endl << endl;
cout << "Res:" << endl << res << endl;
return 0;
}
Then the values in res
are only the values at the indices you need:
Mat:
[0, 1, 2, 3;
4, 5, 6, 7;
8, 9, 10, 11;
12, 13, 14, 15]
Res:
[0, 2;
8, 10]
In order to restore the values to original position, you can use the Kronecker product (not available in OpenCV, but can be easily implemented) with a suitable pattern. This will produce:
Mat:
[0, 1, 2, 3;
4, 5, 6, 7;
8, 9, 10, 11;
12, 13, 14, 15]
Res:
[0, 2;
8, 10]
Res Modified:
[1, 3;
9, 11]
Restored:
[1, 0, 3, 0;
0, 0, 0, 0;
9, 0, 11, 0;
0, 0, 0, 0]
Code:
#include <opencv2/opencv.hpp>
#include <algorithm>
#include <iostream>
using namespace cv;
using namespace std;
Mat kron(const Mat A, const Mat B)
{
CV_Assert(A.channels() == 1 && B.channels() == 1);
Mat1d Ad, Bd;
A.convertTo(Ad, CV_64F);
B.convertTo(Bd, CV_64F);
Mat1d Kd(Ad.rows * Bd.rows, Ad.cols * Bd.cols, 0.0);
for (int ra = 0; ra < Ad.rows; ++ra)
{
for (int ca = 0; ca < Ad.cols; ++ca)
{
Kd(Range(ra*Bd.rows, (ra + 1)*Bd.rows), Range(ca*Bd.cols, (ca + 1)*Bd.cols)) = Bd.mul(Ad(ra, ca));
}
}
Mat K;
Kd.convertTo(K, A.type());
return K;
}
int main(int argc, char **argv)
{
Mat1b mat = (Mat1b(4, 4) << 0, 1, 2, 3,
4, 5, 6, 7,
8, 9, 10, 11,
12, 13, 14, 15);
Mat1b res;
resize(mat, res, Size(0, 0), 0.5, 0.5, INTER_NEAREST);
cout << "Mat:" << endl << mat << endl << endl;
cout << "Res:" << endl << res << endl << endl;
// Work on Res
res += 1;
cout << "Res Modified:" << endl << res << endl << endl;
// Define the pattern
Mat1b pattern = (Mat1b(2,2) << 1, 0,
0, 0);
// Apply Kronecker product
Mat1b restored = kron(res, pattern);
cout << "Restored:" << endl << restored << endl << endl;
return 0;
}