[Octave-bug-tracker] [bug #46745] image package: imresize default method (bicubic) Matlab incompatible for the border pixels

[Carnë Draug]

Update of bug #46745 (project octave):

                 Release:                   3.8.1 => other                  
                 Summary: imresize default method (bicubic) return <0 and >1
values => image package: imresize default method (bicubic) Matlab incompatible
for the border pixels

    _______________________________________________________

Follow-up Comment #2:

With the report as it is, the function seems to behave as expected. Since the
image has values outside the [0 1] range, it is even of an integer class, and
the function is supposed to return a function of the same class as input, the
output image should also have values outside the [0 1] range.

I then tested after converting the image to class double (with im2double):


octave> im = imread ('Example.png');
octave> im = im2double (im);
octave> imr = imresize (im, [512 512]);
octave> max (imr(:))
ans =  1.2466
octave> min (imr(:))
ans = -0.20069


However, this is an artifact of the resizing and only happens on the borders.
I got it tested on Matlab and apparently Matlab also returns values outside [0
1]. They do return different values so I'm guessing they perform some special
padding before interpolation.

    _______________________________________________________

Reply to this item at:

  <http://savannah.gnu.org/bugs/?46745>

_______________________________________________
  Message sent via/by Savannah
  http://savannah.gnu.org/