[Octave-bug-tracker] [bug #60322] [octave forge] (image) imresize bicubic interpolation inaccurate

[Hartmut]

Follow-up Comment #13, bug #60322 (project octave):

I just had a short look at the more formal aspects of this latest patch (from
comment #12). And most things looked fine to me. (Good test coverage, all
tests pass, doc strings, coding style.) One single point I noted:

* It seems to me (from the comments in the tests) that Octave's imresize only
gives Matlab compatible results, when Matlab runs with the anti-aliasing
turned off. If this is true, I think this should be mentioned somewhere in the
help string of imresize. (Is this only relevant when shringing images, as
suggestd by the Matlab help page?)

I would like Avinoam to also have a last look at the algorithmic logic of this
new code. He started to look into this, that's why I hesitate to also dig
deeply into this. Afterwards we should hopefully be ready to push this code
improvement to the image repository.

One other thing that I realised when reading through the history of this bug
report is: a new bug report regarding the incompatibility of the bicubic
method in interp2.m (in core Octave) has not been filed so far (as discussed
in comment #4 and #5). I would be very good to still do this. Or maybe I have
missed this new bug report, then please just mention its bug report number
here.

    _______________________________________________________

Reply to this item at:

  <https://savannah.gnu.org/bugs/?60322>

_______________________________________________
  Message sent via Savannah
  https://savannah.gnu.org/