QA (public) » openQA Project (public)

It was discussed that you will hunt for the issue stating something similar which you expected to be there in our issue tracker and we could not find it. Guess you are just getting old like the all of us :-)

As a side note, the excludes are currently wrong from a algorithm perspective:

The excluded area is overdrawn with a fill color in both the needle and on the screen.

As long as the needle matches at its original location, this works as intended (although
the MSE/similarity value is biased, as the mean value is calculated for the whole area,
not the relevant area).

If the needle is shifted, excluded areas in the scene and the needle no longer line up, and
the result is wrong.

Currently, the tests having exclude areas only match the needle in the original
location, so this misbehaviour is not detected.

Fixing this is actually not to hard:
As the error metric is the (mean of) squared difference, which is linear, first the error
for the whole needle area has to be calculated (which typically gives us a 101x101 size
error matrix), next the same has to be done for every excluded area of the needle (each
resulting in another 101x101 size error matrix). The correct error matrix can be calculated
by subtracting the latter ones from the first matrix, the minimum is the best match as before.