Distance transform : - c++

Question

Kindly help me with the working of Distance transform and rectify the errors. I have tried Borgefors' method which has defined values for Eucledian measure. I get all zeros as output.
Below is the code which i have tried.
int _tmain(int argc, _TCHAR* argv[])
{
Mat v = imread("ref.png", 0);
imshow("input", v);
Mat forward = (Mat_<uchar>(5, 5) << 0, 11, 0, 11, 0, 11, 7, 5, 7, 11, 0, 5, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0);
Mat backward = (Mat_<uchar>(5, 5) << 0,0,0,0,0, 0,0,0,0,0, 0, 0, 0, 5, 0, 11, 7, 5, 7, 11, 0, 11, 0, 11, 0);
Mat op = cv::Mat::zeros(v.size(), CV_32FC1);
cout << forward;
cout << backward;
int r = v.rows;
int c = v.cols;
float min=100, x = 0;
int size = 3;
int lim = size / 2;
int a, b;
for (int i = lim; i <= r-1-lim; i++)
{
for (int j = lim; j <= c-1-lim; j++)
{
for (int k = -lim; k <= lim; k++)
{
for (int l = -lim; l <= lim; l++)
{
a = (v.at<uchar>(i + k, j + l));
b=(forward.at<uchar>(k + lim, l + lim));
x = a + b;
if (x>0 && min> x)
min = x;
}
}
op.at<float>(i, j) = min;
}
}
cout << min;
for (int i = (r-1-lim); i >lim; i--)
{
for (int j = (c-1-lim); j >lim; j--)
{
for (int k = -lim; k <= lim; k++)
{
for (int l = -lim; l <= lim; l++)
{
a = (v.at<uchar>(i + k, j + l));
b = (forward.at<uchar>(k + lim, l + lim));
x = a + b;
if (x >0 && min> x) min = x;
}
}
op.at<float>(i, j) = min;
}
}
cout << op;
Mat res = cv::Mat::ones(v.size(), CV_8UC1);
normalize(op, res, 0, 255, NORM_MINMAX);
imshow("output",res);
waitKey(0);
return 0;
}
Which is the best method and why it is the best way to implement Distance Transform?

Answer 1

Here is how to fix your code:
Apply the backward mask in the backward loop, you apply the same mask there as in the forward loop.
Use only the defined weights, the values in the mask where you wrote 0 are not part of the mask. Those pixels don't have a distance of 0!
As for your second question, it's probably out of scope for SO. But what the best method is depends very much on the goal. You have a fast and relatively accurate method here, there are other methods that are exact but more expensive.