I've seen the same effect when the crop is narrowed down to just the chart,
no additional edges. Next time I run into this I'll post it again.
My samples demonstrate the situation of indeterminate behavior. Please take
me at my word above.
If you imagine a bell curve of matching results, based on the chanciness of
my demo — and I am mosdef not alone as the forum history attests — there
are gonna be a group of users, maybe small, who are just stuck. When I hit
this a week ago, I had to goof around for a half hour to hit a crop that
worked for me.
If scanin produced an error message offering hints, that would help, even
if it were just "sry to you are an unlucky one try next life" that would be
a step ahead of dull failure.
Thanks for taking a look
/w
On Thu, Oct 14, 2021 at 18:48 Graeme Gill <graeme@xxxxxxxxxxxxx> wrote:
Wire ~ wrote:
Attached are links two targets A and B one of which is accepted by scaninyou
and the other isn't... Imagine you have started with A one and
scanin barfs, then you try crops and they don't work and 5 tries later
find B.
Neither will be recognized properly. The second one ends up a line out.
Part of the issue is that because the patches are wedges, the edge
contrast isn't
very good, and the grain noise is roughly similar in amplitude to the
edges.
The primary issue is all the writing below the chart ("ISO 12641-2 ...
Lasersoft.. etc.)
adds a lot of extraneous edges that confuses the recognition since it isn't
structured as edges, and isn't expected in the recognition.
Crop that out or paint it over with grey, and both charts are recognized.
i.e. it only needs to be roughly cropped, but it helps to crop out anything
extraneous.
Cheers,
Graeme.
