rotate makes photo fuzzy
Mark Mcknight
When rotating a photo a couple of degrees using transform tools it seems to
de-focus the photo. It doesn't occur when I rotate the photo using the
images | transform menu (the one where there are three rotate choices: 90,
180 and 270 degrees).
Anyone else seen this and know of an option for minor adjustments to
rotation or know of a way to fix the damage? I've tried, but can't get the
focus back to an acceptable level.
Thanks, Mark
Jean-Louis HAMEL
Mark Mcknight wrote:
>
> Anyone else seen this and know of an option for minor adjustments to
> rotation or know of a way to fix the damage? I've tried, but can't get the
> focus back to an acceptable level.
> Thanks, Mark
>
In the File menu of the toolbox clic on Preferences/Environment and set
the cubic interpolation. This make transformations slower but the damage
is minimized.
JLH
Leonard Evens
Mcknight" <mark-m...@home.com> wrote:
> Using suse 7.3, Gimp 1.2*
> When rotating a photo a couple of degrees using transform tools it seems
> to de-focus the photo. It doesn't occur when I rotate the photo using
> the images | transform menu (the one where there are three rotate
> choices: 90, 180 and 270 degrees).
Rotating through multiples of 90 degrees doesn't require any
interpolation. Each pixel is just rotated the appropriate amount and
placed in the appropriate position.
> Anyone else seen this and know of an option for minor adjustments to
> rotation or know of a way to fix the damage? I've tried, but can't get
> the focus back to an acceptable level.
I agree that setting cubic interpolation in preferences will minimize
this effect.
I've been scanning medium format negatives with an Epson 1640 Photo
flatbed scanner, and manipulating them in the Gimp. I often use the
transform tool to rotate or to correct perspective distortion. I have
noticed that it does have a slight softening effect, something like
applying a Gaussian blur with a radius of 1 or 2. But I can usually
recover anything that was lost by applying some sharpening. Often I
don't bother because my previous sharpening has accentuated "graininess"
which the rotation then softens slightly without affecting larger scale
sharpness.
But my scans produce images 3500 x 3500 or larger. I imagine the effect
might be more objectionable with smaller scans.
> Thanks, Mark
--
Leonard Evens l...@math.northwestern.edu 847-491-5537
Dept. of Mathematics, Northwestern Univ., Evanston, IL 60208
Mark Mcknight
> In article <X7wJ7.95526$Gh2.30...@news2.rdc1.bc.home.com>, "Mark
> Mcknight" <mark-m...@home.com> wrote:
>
>
> Rotating through multiples of 90 degrees doesn't require any
> interpolation. Each pixel is just rotated the appropriate amount and
> placed in the appropriate position.
>
>> Anyone else seen this and know of an option for minor adjustments to
>> rotation or know of a way to fix the damage? I've tried, but can't get
>> the focus back to an acceptable level.
>
> I agree that setting cubic interpolation in preferences will minimize
> this effect.
Changing from linear to cubic improved the quality in my application.
> I've been scanning medium format negatives with an Epson 1640 Photo
> flatbed scanner, and manipulating them in the Gimp. I often use the
> transform tool to rotate or to correct perspective distortion. I have
> noticed that it does have a slight softening effect, something like
> applying a Gaussian blur with a radius of 1 or 2. But I can usually
> recover anything that was lost by applying some sharpening. Often I
> don't bother because my previous sharpening has accentuated "graininess"
> which the rotation then softens slightly without affecting larger scale
> sharpness.
>
> But my scans produce images 3500 x 3500 or larger. I imagine the effect
> might be more objectionable with smaller scans.
That's interesting, makes sense when I think about it. I was working on
some 600 dpi photos for the web.
I use a HP Photosmart scanner that can do 2400 dpi. I'll try it at that
resolution.
Mark
David Hodson
> [...] I often use the
> transform tool to rotate or to correct perspective distortion. I have
> noticed that it does have a slight softening effect, something like
> applying a Gaussian blur with a radius of 1 or 2. But I can usually
> recover anything that was lost by applying some sharpening.
Since I was the last person (I think) to modify the transformation
resampling code ... the current values are the best compromise that
I could find between blur and ringing. If you decrease the blur, you
will get ringing on sharp edges. As you point out, it's always possible
to sharpen after the transform (which probably gives the same result
as changing the convolution kernel, although I'd hate to try to prove
that mathematically).
Of course, it's important not to do multiple transformations on the
same image. Maybe in 1.3, someone (me?) could modify the transform tool
so that it lets you specify a series of transformation steps, but only
performs a single image processing step. (I don't think it can do this
yet, but I haven't checked.)
--
David Hodson -- this night wounds time
Laurent Despeyroux
>
> Since I was the last person (I think) to modify the transformation
> resampling code ... the current values are the best compromise that
> I could find between blur and ringing. If you decrease the blur, you
Just a question : what is ringing ? Can you give a URL where we can
see that ?
--
Laurent Despeyroux
Damaeus
I've never tried rotating images in Gimp, but I know that I was creating
frames for an animated GIF in which I wanted to do an animation of an
image rotating 180 degrees in 19 frames, 10 degrees for each frame. By
the time I'd rotated the image 19 times, it was unacceptably fuzzy, like I
was seeing it through frosted glass almost.
Damaeus
Leonard Evens
<hod...@ozemail.com.au> wrote:
I do multiple transformations on my images regularly without any obvious
degradation of the image. So your code may be better than you think it
is. But as I mentioned previously my scans have fairly high pixel
density.
Olivier Ripoll
(google is your friend)
I've found a nice (it uses wilber :) ) URL about image processing and
iterpolation here:
http://www.engineering.uiowa.edu/~gec/248_s00_students/blake_carlson/hw2/
and about ringing, here is an explanation at mathworks (the company of
matlab) :
http://www.mathworks.com/access/helpdesk/help/toolbox/images/deblur10.shtml#9482
Sincerely,
Olivier Ripoll.
James Waldby
> Mark Mcknight ... posted:
...
> > When rotating a photo a couple of degrees using transform tools it seems to
> > de-focus the photo.
> I've never tried rotating images in Gimp, but I know that I was creating
> frames for an animated GIF in which I wanted to do an animation of an
> image rotating 180 degrees in 19 frames, 10 degrees for each frame. By
> the time I'd rotated the image 19 times, it was unacceptably fuzzy, like I
> was seeing it through frosted glass almost.
To avoid that problem, only rotate the original image, by
10, 20, 30 ... degrees, rather than rotating images 10
degrees sequentially.
James Waldby
...
> Of course, it's important not to do multiple transformations on the
> same image. Maybe in 1.3, someone (me?) could modify the transform tool
> so that it lets you specify a series of transformation steps, but only
> performs a single image processing step. (I don't think it can do this
> yet, but I haven't checked.)
Do you mean you would build up a single composite transformation
matrix for a series of affine transformations, and then apply
the composite matrix once rather than applying the intermediate
matrices? That would lead to less image degradation, but for
it to be usable there needs to be a way to record and edit
sequences of transformations. That is, while finding out what
sequence you will need to apply, you may need to see intermediate
images (for example, perspectives applied at several corners),
but for image quality would want to apply the composite sequence
to the original image.
Martin Schulz
> Do you mean you would build up a single composite transformation
> matrix for a series of affine transformations, and then apply
> the composite matrix once rather than applying the intermediate
> matrices? That would lead to less image degradation, but for
> it to be usable there needs to be a way to record and edit
> sequences of transformations.
This should work if there is nothing else than rotations or
translations. Ohter actions such as combinations with other layers may
not fit into that framework.
My understanding is that gimp is strongly built on that step-by-step
assumption, therefore I see certain difficulties. Clearly, consecutive
series of rotations and/or translations can and should be fused into a
single one to prevent unnecessary degradation of the picture.
Yours,
--
Martin Schulz sch...@iwrmm.math.uni-karlsruhe.de
Uni Karlsruhe, Institut f. wissenschaftliches Rechnen u. math. Modellbildung
Engesser Str. 6, 76128 Karlsruhe
David Hodson
>
> David Hodson wrote:
> > [...] Maybe in 1.3, someone (me?) could modify the transform tool
> > so that it lets you specify a series of transformation steps, but only
> > performs a single image processing step.
> Do you mean you would build up a single composite transformation
> matrix for a series of affine transformations, and then apply
> the composite matrix once rather than applying the intermediate
> matrices?
You know how the controls switch between scale, rotate, etc.? I was
thinking that you should be able to (say) enable scale, stretch the
grid, then switch to rotate and rotate the scaled grid, before applying
the transform. Just a UI thing, but seems like it might be useful.
Gunnar Koppel
On Sat, 17 Nov 2001, Jean-Louis HAMEL wrote:
> In the File menu of the toolbox clic on Preferences/Environment and set
> the cubic interpolation. This make transformations slower but the damage
> is minimized.
Thank you! This one was very needed here :)
But even this linear interpolation was much slower than i had experience
with Photoshop 3.0.5 (Pentium 200 / 64 MB / Win 95). Now i use Gimp 1.2 on
Pentium II Celereon 466 with 192 MB RAM and this is much slower. Why? It
is big difference in algorithms?
Alati Teie
Wanradt Koell