This next part of code does exactly what I want it to do. dem_rows and dem_cols contain float values for a number of things i can identify in an image, but i need to get the nearest pixel for each of them, and than to make sure I only get the unique points, and no duplicates.

The problem is that this code is ugly and as far as I get it, as unpythonic as it gets. If there would be a pure-numpy-solution (without for-loops) that would be even better.

# next part is to make sure that we get the rounding done correctly, and than to get the integer part out of it 
# without the annoying floatingpoint-error, and without duplicates
fielddic={}
for i in range(len(dem_rows)):
    # here comes the ugly part: abusing the fact that i overwrite dictionary keys if I get duplicates
    fielddic[int(round(dem_rows[i]) + 0.1), int(round(dem_cols[i]) + 0.1)] = None
# also very ugly: to make two arrays of integers out of the first and second part of the keys 
field_rows = numpy.zeros((len(fielddic.keys())), int)
field_cols = numpy.zeros((len(fielddic.keys())), int)
for i, (r, c) in enumerate(fielddic.keys()):
    field_rows[i] = r
    field_cols[i] = c