I have to find an maximum in an Octave with a 3x3x3 neighborhood. That means I have four layers on top of each other and between this layers i have to find a maxima. For illustration here an image. That is not what Iam doing but it represent what is my issue.
Octave layer image http://docs.opencv.org/trunk/_images/sift_dog.jpg
Now for the maximum surpression I found thi paper: Efficient Non-Maximum Suppression. Here is an fast way explained to find a maxima in an Image. This is only the 2D case but it shouldn't be a problem to shift this in 3D space. My problem now is really the pseudo code understanding. I have the pseudo-code:
The problem is the red marked part. There I have for loop but I have no idea how to apply the "-[i, i+n] x [j, j+n]" to the loop. That is the moment solution:
//find local maxima after paper implementation not finished yet
private Vector<Integer> FindLocalMaximum(Image image)
{
Vector<Integer> list = new Vector<Integer>();
int n = 1;
int step = 2*n + 1;
for(int i = n; i < image.GetWidth()-n; i =step)
for(int j = n; j < image.GetHeight()-n; j =step)
{
int mi = i;
int mj = j;
for(int i2 = i; i2 < i + n; i2++ )
for(int j2 = j; j2 < j + n; j2++ )
if(image.GetPixel(i2, j2) > image.GetPixel(mi, mj))
{
mi = i2;
mj = j2;
}
boolean found = true;
failed:
for(int i2 = mi - n; i2 < mi + n; i2++ )
for(int j2 = mj - n; j2 < mj + n; j2++ )
if(image.GetPixel(i2, j2) > image.GetPixel(mi, mj))
{
found = false;
break failed;
}
if(found)
{
int pos = mj * image.GetWidth() + mi;
list.add(pos);
}
}
return list;
}
So how surprise it doesn't work. Has somebody an idea what I have to do at the red marked part.
I'll give you an example in pseudocode:
listA = [1, 2, 3]
listB = [a, b, c]
listA x listB = [(1, a), (1, b), (1, c), ...]
# excluded
listAe = [1, 3]
listBe = [a, b]
listAe x listBe = [(1, a), (1, b), ...]
# result
listA x listB - listAe x listBe = [(1, c), (2, a), (2, b), (2, c), (3, c)]
Now you should only iterate over result pairs.