Tuning the Detection
The defaults find something in most images. This page is how to make them find the right thing.
The settings that matter
Only three, which is much of the appeal:
method-specific/fixed/level-
The brightness a pixel must exceed to count as part of a region. Commission it with Analysing a Scene rather than by guessing - it suggests a level and shows you the histogram the level has to sit in.
detect-dark-
Turn on when the parts are darker than the background.
fill-holes-
Turn on to measure a region as the shape its outline encloses. A washer measures as a ring with this off and as a disc with it on, and both are true answers.
There is no area window here. Every region is reported, and dropping the ones too small or too large to be a part is done where the result is read - only there is it known what a part is.
Preprocessing
Try preprocess/method when regions have soft or noisy edges. LOG and DOG both emphasise features
of a particular size - set the sigmas to roughly the size of what you are looking for - and DOH
responds to blob-like structure. All three normalise their output, so the intensity thresholds above
may need revisiting after enabling one.
When no level works
If the reading will not settle, check system/analyse-scene/output/area-sensitivity before touching
the settings. Above 0.05 the picture is brittle — one grey level costs 5 % of the measured area —
and no value in input/ repairs that. It is fixed at the camera: sharper edges, more contrast, or
more pixels on the part. See
Analysing a Scene for which of the three it is.