[sane-devel] sanei_magic

[m. allan noah]

Historically, many of us (myself included) have advocated for software
features like deskew, despeckle and auto-crop to be done in the
front-end. This has obvious advantages for code reuse and
maintainability, since there appear to be fewer front-ends than
backends. But, I think the biggest reason is more selfish. These
features are hard to write, and I, for one, would rather that it was
someone else's problem :)

However, many current brands of scanner can do various image
processing steps in hardware, but only on the medium & high cost
machines. Users of the low cost machines get these features via
software in the vendor's proprietary drivers, and don't see any
difference. Also, the number of front-ends is a bit of a red herring,
since the most popular ones by far are scanimage, scanadf, and Xsane,
none of which has yet provided these features. Besides, the first two
of these are produced by us, so much for passing the workload to
someone else :)

So, in the interest of consistency for all users, I have undertaken to
provide these features somewhere in sane. I considered adding this
type of code to scanimage, but then the users of Xsane cannot reach
it. Then I started writing a 'middle-end', but after I got done
replicating much of the logic in the dll backend in order to load
lower layers backends, I switched to adding the code to the backends
directly. Such code has been in the canon_dr backend for a few weeks.
Though simplistic, it is fairly well tested, and does work reasonably.

Now, to improve reusability, I have spun a more generic version of
that code into sanei_magic, updated sane-fujitsu to use it, and
committed both to git.

Details:
1. Deskew, autocrop and despeckle
2. Deskew and autocrop are each divided into two functions, one to
find the appropriate boundaries/angles, and one to correct the image
based on those parameters. This gives the caller a chance to inspect
and/or change the outcome of the operation.
3. Despeckle takes a maximum diameter to remove, and fills the area
with average surrounding color
4. Deskew uses a simplified iterative 'original' Hough transform which
searches for the top and left edges of the media inside the image.
This is less sensitive to false positives than looking at content, but
does not function well if no media edge is visible. I continue to
search for alternative mechanisms.

With all this code I am certainly open to modifications and
adjustments (even renaming, sanei_magic was the best I could do). I
will even accept criticism for even including it in the first place,
though I think once users take notice of it, these will become popular
features.

allan
-- 
"The truth is an offense, but not a sin"