Shrinking a picture --- and it becomes blur !!
hsy...@gmail.com
I do computer graphics as a hobby, and have produced quite a number of
stunning graphics. Often time though, when I shrink the graphic to put
them online, they become blurred !!
I do all kinds of computer graphics, from fractals to virtual
landscape, to sci-fi rendering, using softwares ranging from photoshop
to terragen to povray.
When I am satisfied with a certain creation, I often make a master
copy with the resolution of 8192 X 6144 pixel. Why that size? Because
that's the largest size my puny computer (dual-core 3GHz CPU running
XP with 4GB RAM) can produce within a reasonably timeframe. (Give or
take 8 hours for rendering).
As the filesize for a JPG with 8192 X 6144 resolution may go up to 30+
MB, I often have to shrink them to a more reasonable 1024 X 768,
filesize about 800 KB or so.
However, I found that when I do that, many interesting minute details
that were in the 8K X 6K pictures (even when I shrink fit it to my
1024X768 desktop as wallpaper) are GONE. In the 1024 X 768 JPG files,
all those details become blurred. No matter it's a JPG ---> JPG
shrink, or BMP ---> JPG shrink, or TIFF ---> JPG shrink, all those
details are GONE !!
I have experimented with many different graphic / photo softwares in
the shrinking process, all of them give me the same "blurring" effect.
Now my questions to all you Gurus as below ---
1. Can you tell me of the best way to shrink a 8192X6144 size graphic
to
1024X768 size graphic without losing the interesting details?
2. Which software do you recommend to carry out the shrinking
operation?
Thank you all in advance !!!
Sincerely,
Lee
N
news:52088436-f3ae-42a8...@q24g2000prf.googlegroups.com...
What were expecting? If you remove seven eighths of the data then seven
eighths of the data will be gone.
Andrew Morton
> When I am satisfied with a certain creation, I often make a master
> copy with the resolution of 8192 X 6144 pixel. Why that size? Because
> that's the largest size my puny computer (dual-core 3GHz CPU running
> XP with 4GB RAM) can produce within a reasonably timeframe. (Give or
> take 8 hours for rendering).
You should render it to the size you're going to use. And then it'll only
take a few minutes to render. You can always render it to a different size
if needed.
<snip: saves as jpeg>
> However, I found that when I do that, many interesting minute details
> that were in the 8K X 6K pictures (even when I shrink fit it to my
> 1024X768 desktop as wallpaper) are GONE.
That's what jpeg compression does. It is a lossy compression method.
How about saving as a png? It may not produce files as small as jpeg, but
the images will not be blurred (assuming you don't use jpeg compression in
the png!).
Andrew
mark.t...@gmail.com
As advised, rendering to the size you want may be the best way.
Or try the "Lanczos" algorithm - eg in Irfanview. Also, downsizing in
steps may work better, in other words, try reducing in steps of say
20% (or even less), and then experiment with light sharpening (USM) at
each step - Irfanview has this function built in, but I can usually do
a little better manually. I've found what works for some images,
doesn't work as well for others.. and I've never experimented with
rendered images, so all this may be useless... (O:
The best you can hope for is one/two-pixel sharpness, so maybe you are
expecting too much?
bugbear
> On Apr 29, 4:39 pm, hsyq...@gmail.com wrote:
>>
>> As the filesize for a JPG with 8192 X 6144 resolution may go up to 30+
>> MB, I often have to shrink them to a more reasonable 1024 X 768,
>> filesize about 800 KB or so.
>
>
> Or try the "Lanczos" algorithm
Since the downsize is an exact integer (factor of 8)
I'm not sure Lanczos (or anything "better" than bilinear)
would help.
BugBear
pen...@freemail.c3.hu
Interesting problem you got there. Perhaps you could speed up the
rendering time by getting some fancy video card that is powered by
super-duper ultra fast GPU ??
Which GPU the best? Hmmm.........
k
|
| I do computer graphics as a hobby, and have produced quite a number of
| stunning graphics. Often time though, when I shrink the graphic to put
| them online, they become blurred !!
| 2. Which software do you recommend to carry out the shrinking
| operation?
anything with the right algorithm, and the algoriothm makes all the
difference
..and downsizing is quite, quite different from upsizing!
a basic primer on algorithm for resizing can be found here: (grr.. the web
site's gone!)
http://web.archive.org/web/20060409125805/http://www.interpolatethis.com/int
erp.html
and some comparisons
http://www.cambridgeincolour.com/tutorials/image-resize-for-web.htm
http://www.cambridgeincolour.com/tutorials/digital-photo-enlargement.htm
mark.t...@gmail.com
But I did also refer to downsizing in much smaller steps - and Lanczos
will possibly help there...
Allen
machine, taking up memory and cycles and competing with your graphics
program? If you don't know, press CTRL-ALT-DEL to bring up Windows Task
Manager and click on Applications, the Processes, then Performance. You
might be forcing your graphics program to plod through knee-deep mud to
do its job.
Allen
Martin Brown
> On Mon, 28 Apr 2008 23:39:48 -0700 (PDT), hsy...@gmail.com wrote:
>> snip
>> Now my questions to all you Gurus as below ---
>>
>> 1. Can you tell me of the best way to shrink a 8192X6144 size graphic
>> to
>> 1024X768 size graphic without losing the interesting details?
Render the image to the size you want it in the first place. Then choose
your JPEG compression parameters and chroma subsampling carefully and
wisely to get the best size vs quality tradeoff. Several packages allow
you to see the size and preview artefacts against the original.
Before any meaningful answer about downsampling is possible you are
going to have to answer the question "what do I mean by an interesting
detail".
>>
>> 2. Which software do you recommend to carry out the shrinking
>> operation?
>>
>> Thank you all in advance !!!
>
> IrfanView gives you a choice of 5 filters when resizing. Try each of these
> until you get the desired results.
> http://www.irfanview.com
Downsampling by a factor of 8 in each linear dimension will necessarily
have a pretty severe effect on the maximum spatial frequency of the
original image that can be retained. There is no free lunch.
Regards,
Martin Brown
** Posted from http://www.teranews.com **
Marco Tedaldi
> Or try the "Lanczos" algorithm - eg in Irfanview. Also, downsizing in
> steps may work better, in other words, try reducing in steps of say
> 20% (or even less), and then experiment with light sharpening (USM) at
> each step - Irfanview has this function built in, but I can usually do
> a little better manually. I've found what works for some images,
> doesn't work as well for others.. and I've never experimented with
> rendered images, so all this may be useless... (O:
>
Can you explain, why there should be any advantage in scaling down an
image in smaller steps? I've read this several times by now and can just
say that this seems like some quite big bullshit to me.
Maybe you can prove me wrong by supplying three images:
- original
- downsized in one step
- downsized in several steps
and the exact steps you used to get this result.
The logic behind downsizing in several stps is not clear to me. you
downsize by 20%... so you lose some information and the computer has to
interpolate ("guess" some values). After this you use this guessed
values to further shrink the image... and this should lead to better
results (and more detail) than downsizing in one step?
If this was true, I'd fire ALL (every single one) developer of the
scaling-procedures in every graphics program!
And think a bit further: IF there was an advantage in scaling down in
steps, the would be such a function built in the graphics-applications.
Ok, you can prove me wrong. Good luck! Mathematics and
Information-Theory is normally hard to beat!
> The best you can hope for is one/two-pixel sharpness, so maybe you are
> expecting too much?
That's for sure. Maybe he would be happiest with the cheapest
scaling-procedure available. No interpolation at all (i.e. nearest
neighbor). This will give him the sharpest image possible.
Marco
--
Dimage A2, Agfa isolette
http://flickr.com/photos/kruemi
http://profile.imageshack.us/user/kruemi/images
Marco Tedaldi
> Or try the "Lanczos" algorithm - eg in Irfanview. Also, downsizing in
> steps may work better, in other words, try reducing in steps of say
> 20% (or even less), and then experiment with light sharpening (USM) at
> each step - Irfanview has this function built in, but I can usually do
> a little better manually. I've found what works for some images,
> doesn't work as well for others.. and I've never experimented with
> rendered images, so all this may be useless... (O:
>
Can you explain, why there should be any advantage in scaling down an
image in smaller steps? I've read this several times by now and can just
say that this seems like some quite big bullshit to me.
Maybe you can prove me wrong by supplying three images:
- original
- downsized in one step
- downsized in several steps
and the exact steps you used to get this result.
The logic behind downsizing in several stps is not clear to me. you
downsize by 20%... so you lose some information and the computer has to
interpolate ("guess" some values). After this you use this guessed
values to further shrink the image... and this should lead to better
results (and more detail) than downsizing in one step?
If this was true, I'd fire ALL (every single one) developer of the
scaling-procedures in every graphics program!
And think a bit further: IF there was an advantage in scaling down in
steps, the would be such a function built in the graphics-applications.
Ok, you can prove me wrong. Good luck! Mathematics and
Information-Theory is normally hard to beat!
> The best you can hope for is one/two-pixel sharpness, so maybe you are
> expecting too much?
That's for sure. Maybe he would be happiest with the cheapest
aruzinsky
Instead of excess verbosity, you should post links to crops of before
and after images showing the problem areas.
You (plural) are not entitled to ignore the adage, "A picture is worth
a thousand words."
KatWoman
are you setting to bicubic sharper??
I heard this is a better option to tick for sizing down
what if your original is not jpeg (compressed) but instead saved as tiff
uncompressed?
then you are not compressing an already compressed format
are you choosing high or maximum for quality of the downsized jpg?
Dave Martindale
>> Or try the "Lanczos" algorithm
>Since the downsize is an exact integer (factor of 8)
>I'm not sure Lanczos (or anything "better" than bilinear)
>would help.
The fact that the downsize ratio is an integer factor means it's easy to use a box filter
for the downsizing and get "not bad" results. That doesn't mean Lanczos would not be better
yet. A box filter still produces a fair amount of aliasing due to letting through input
image frequency components that are above the Nyquist limit for the output resolution.
Lanczos is better at attenuating these frequencies and suffers less aliasing artifacts. At
the same time, it's better at retaining frequencies just below Nyquist that a box filter
attenuates more than necessary.
Now, a reasonable-sized Lanczos downsampling filter will cost you more time than a box
filter - but it should be small compared to the original render time.
Dave
tacit
<52088436-f3ae-42a8...@q24g2000prf.googlegroups.com>,
hsy...@gmail.com wrote:
> I do computer graphics as a hobby, and have produced quite a number of
> stunning graphics. Often time though, when I shrink the graphic to put
> them online, they become blurred !!
Well, of course! What did you expect?
If you resize an image to make it smaller, you remove pixels. When you
remove pixels, you remove detail.
On top of that, if you save as JPEG, you remove even more detail. JPEG
compression is "lossy." It was invented for situations where file size
is critical and image quality is not important. It makes files smaller
by discarding image detail.
--
Photography, kink, polyamory, shareware, and more: all at
http://www.xeromag.com/franklin.html
bugbear
> In article
> <52088436-f3ae-42a8...@q24g2000prf.googlegroups.com>,
> hsy...@gmail.com wrote:
>
>> I do computer graphics as a hobby, and have produced quite a number of
>> stunning graphics. Often time though, when I shrink the graphic to put
>> them online, they become blurred !!
>
> Well, of course! What did you expect?
>
> If you resize an image to make it smaller, you remove pixels. When you
> remove pixels, you remove detail.
>
> On top of that, if you save as JPEG, you remove even more detail. JPEG
> compression is "lossy." It was invented for situations where file size
> is critical and image quality is not important. It makes files smaller
> by discarding image detail.
>
On rereading the OP's post, I think the
(interesting) discussion on sampling algorithms
is besides the point.
The key phrase (I think) is:
> However, I found that when I do that, many interesting minute details
> that were in the 8K X 6K pictures (even when I shrink fit it to my
> 1024X768 desktop as wallpaper) are GONE.
I think this implies that simply setting the "big image"
as desktop gives results the OP finds acceptable, which I imagine
involves pretty crude down sampling.
Which leaves JPEG artifacts as the culprit.
BugBear
Chris Malcolm
> On Apr 29, 8:13?pm, bugbear <bugbear@trim_papermule.co.uk_trim> wrote:
>> mark.thoma...@gmail.com wrote:
>> > On Apr 29, 4:39 pm, hsyq...@gmail.com wrote:
>>
>> >> As the filesize for a JPG with 8192 X 6144 resolution may go up to 30+
>> >> MB, I often have to shrink them to a more reasonable 1024 X 768,
>> >> filesize about 800 KB or so.
>>
>> > As advised, rendering to the size you want may be the best way.
>>
>> > Or try the "Lanczos" algorithm
>>
>> Since the downsize is an exact integer (factor of 8)
>> I'm not sure Lanczos (or anything "better" than bilinear)
>> would help.
>>
> But I did also refer to downsizing in much smaller steps - and Lanczos
> will possibly help there...
I experimented with downsizing in Irfanview with the Lanczos option
and found that doing it in a number of steps never produced better
results, and sometimes produced worse ones.
--
Chris Malcolm c...@infirmatics.ed.ac.uk DoD #205
IPAB, Informatics, JCMB, King's Buildings, Edinburgh, EH9 3JZ, UK
[http://www.dai.ed.ac.uk/homes/cam/]
Pico
kind, then consider making your important detail in a vector drawing
program. If you can and if it works visually. Vectors will display as the
very best possible given the resolution of the display in use. (Look at US
paper currency under magnification. The engraving could be vectors.)
I have not read nor seen anyone 'mixing' a vector image after scaling with a
scaled raster image. The outcome will be a raster image with punctuations of
very fine lines of detail and even colors (where the colors are unitary).
(Didn't Steve Jobs' R&D people have postscript display in their neXt OS?)
What an idea. I will explore it beginning tomorrow. My current grant runs
out at 5PM today and I will be free for a week or so to use all this neat
stuff for pure play. I mean research.
Pico
> Instead of excess verbosity, you should post links to crops of before
>and after images showing the problem areas.
ZING!
Beryl
> <hsy...@gmail.com> wrote:
...
>> 1. Can you tell me of the best way to shrink a 8192X6144 size graphic
>> to
>> 1024X768 size graphic without losing the interesting details?
...
> What were expecting? If you remove seven eighths of the data then seven
> eighths of the data will be gone.
Much less than 1/8 of the data remains.
(1024 X 768) / (8192 X 6144) = fraction remaining
786432 / 50331648 = 0.015625
Allen
pounds of potatoes. The bag was too heavy, so I took out nine pounds and
threw them away. When I got home I couldn't believe it when I could find
only one pound. I asked my mother how I could get them back, and she
told me to go back to where I threw them away and bring them back;
sadly, I couldn't find them.
Allen
Dave
Stupid..! You should have kept the ten pounds with the one pound
and carry all 11 pounds. You should never pick on a digital career
when you grow up.
(clever people never throw potatoes away, anyway.)
nospam
> (Didn't Steve Jobs' R&D people have postscript display in their neXt OS?)
nextstep/openstep used display postscript and os x uses pdf.
hsy...@gmail.com
3 example online:
A. www.PenangA1.com/png/1K.PNG
B. www.PenangA1.com/png/4K.PNG
C. www.PenangA2.com/png/ORI.PNG
The three examples above are fragments of a drawing that terragen
produced just a few days ago.
All three pictures were taken by the same screen capture program.
If you take a look at Picture C (ori.png), it's from the original
drawing ( resolution: 4096 X 3072 ), with file size of 37MB, in BMP
format. In Picture C you can see patches of green leaves distinctly,
arising from the brown wall.
Picture B represents a screenshot fragment of the 37MB drawing as my
desktop wallpaper (1024 X 768). As you can see from Picture B,
although much smaller than Picture C, the patches of green leaves are
still separated from the brown wall.
However, if you look at Picture A ... the patches of green leaves
kinda melt into the brown wall behind it. Picture A was from a
fragment of a 1024 X 768 picture (JPG format) that I shrunk from the
original 37 MB BMP drawing. When I shrunk it, I use 100% JPG quality,
with the "Lanczos" option.
The most important thing is that comparing Picture A with Picture
B ... as you can see, even if Picture A and Picture B were obtained
from pictures with the same dimension ( 1024 X 768 ), albeit different
pictures, the green leaf patches of Picture B can still be clearly
seen, while Picture A, the green leaves and brown wall are all mixed
up.
Any comment ??
pg
> Sorry for this late reply.
>
> 3 example online:
>
> A.www.PenangA1.com/png/1K.PNG
>
> B.www.PenangA1.com/png/4K.PNG
>
> C.www.PenangA2.com/png/ORI.PNG
Picture C should be www.PenangA1.com/png/ORI.PNG
aruzinsky
>
> - Show quoted text -
Apparently the links are broken. Also, please, don't post BMP, post
small JPEG 100% crops of a problem area because I use a dial up modem.
pg
All are PNG files. Yes, the links are damn broken !!!
Dave
>Apparently the links are broken. Also, please, don't post BMP, post
>small JPEG 100% crops of a problem area because I use a dial up modem.
and
>All are PNG files. Yes, the links are damn broken !!
Why don't you idiots crop the crap?
aruzinsky
http://www.general-cathexis.com/images/ORI_0.125X_LanczosUnExpandedKernel.png
, to this,
http://www.general-cathexis.com/images/ORI_0.125X_LanczosExpandedKernel.png
, then I suggest that you use an unexpanded interpolation kernel.
Typically, commercial softwares expand the interpolation kernel for
reductions because such expansion automatically provides
antialiasing. SAR Image Processor gives the option of not expanding
the kernel. This is undesirable for natural images.
aruzinsky
using an unexpanded kernel is just plain decimation and the kind of
interpolation kernel doesn't matter.
On May 1, 3:49 pm, aruzinsky <aruzin...@general-cathexis.com> wrote:
> If you prefer this,
>
> http://www.general-cathexis.com/images/ORI_0.125X_LanczosUnExpandedKe...
>
> , to this,
>
> http://www.general-cathexis.com/images/ORI_0.125X_LanczosExpandedKern...
> > - Show quoted text -- Hide quoted text -
Martin Leese
Sun's NeWS also used PostScript. All a bit
moot now. These days you would use SVG.
--
Regards,
Martin Leese
E-mail: ple...@see.Web.for.e-mail.INVALID
Web: http://members.tripod.com/martin_leese/
Martin Brown
> Sorry for this late reply.
>
> 3 example online:
>
> A. www.PenangA1.com/png/1K.PNG
>
> B. www.PenangA1.com/png/4K.PNG
>
> C. www.PenangA2.com/png/ORI.PNG
All three time out from here long before they download.
> If you take a look at Picture C (ori.png), it's from the original
> drawing ( resolution: 4096 X 3072 ), with file size of 37MB, in BMP
> format. In Picture C you can see patches of green leaves distinctly,
> arising from the brown wall.
>
> Picture B represents a screenshot fragment of the 37MB drawing as my
> desktop wallpaper (1024 X 768). As you can see from Picture B,
> although much smaller than Picture C, the patches of green leaves are
> still separated from the brown wall.
>
> However, if you look at Picture A ... the patches of green leaves
> kinda melt into the brown wall behind it. Picture A was from a
> fragment of a 1024 X 768 picture (JPG format) that I shrunk from the
> original 37 MB BMP drawing. When I shrunk it, I use 100% JPG quality,
> with the "Lanczos" option.
> Any comment ??
Which JPEG encoder did you use?
And more critically what chroma subsampling - the default used for
photographs does tend to compromise line art images.
If you override the chroma subsampling to save YCrCb with 1,1,1 ratio
then you may get better fidelity (although a larger filesize).
Fundamentally you cannot get around the fact that when the image is
downsampled the new image pixels get bigger and contain colours that are
some weighted average of the original pixels.
pg
Would anyone kindly tell me which software provides the choice of
"expanded" and "unexpanded" kernels of the Lanczos routine?
As I am not a pro in this, while I do own (LEGAL) copies of several
commercial software, I do not recall, photoshop, for instant, offer
the "expanded" and the "unexpanded" kernel versions of Lanczos.
Thank you again !!!
On May 1, 2:49 pm, aruzinsky <aruzin...@general-cathexis.com> wrote:
> If you prefer this,
>
> http://www.general-cathexis.com/images/ORI_0.125X_LanczosUnExpandedKe...
>
> , to this,
>
> http://www.general-cathexis.com/images/ORI_0.125X_LanczosExpandedKern...
>
Denise
news:654686df-9016-407f...@w1g2000prd.googlegroups
.com:
>On May 1, 2:49 pm, aruzinsky <aruzin...@general-cathexis.com>
wrote:
>> If you prefer this,
>>
>> http://www.general-
cathexis.com/images/ORI_0.125X_LanczosUnExpandedKe...
>>
>> , to this,
>>
>> http://www.general-
cathexis.com/images/ORI_0.125X_LanczosExpandedKern...
>>
>> , then I suggest that you use an unexpanded interpolation
kernel.
>>
>> Typically, commercial softwares expand the interpolation kernel
for
>> reductions because such expansion automatically provides
>> antialiasing. SAR Image Processor gives the option of not
expanding
>> the kernel. This is undesirable for natural images.
> Thank you for the info.
>
> Would anyone kindly tell me which software provides the choice
> of "expanded" and "unexpanded" kernels of the Lanczos routine?
>
> As I am not a pro in this, while I do own (LEGAL) copies of
> several commercial software, I do not recall, photoshop, for
> instant, offer the "expanded" and the "unexpanded" kernel
> versions of Lanczos.
>
> Thank you again !!!
Appreciate bottom posting.
Adding:
adobe.photoshop.windows
alt.graphics.photoshop
Denise
> Thank you for the info.
You are a spammer:
http://groups.google.com/groups?as_q=&num=100&scoring=d&hl=en&as_ep
q=&as_oq=&as_eq=&as_ugroup=&as_usubject=&as_uauthors=penang%40catho
lic.org&lr=lang_en&as_drrb=q&as_qdr=&as_mind=1&as_minm=1&as_miny=19
81&as_maxd=2&as_maxm=5&as_maxy=2008&safe=off
aruzinsky
free demo.
> > > - Show quoted text -- Hide quoted text -
Paul Furman
>> Hello !
>>
>> I do computer graphics as a hobby, and have produced quite a number of
>> stunning graphics. Often time though, when I shrink the graphic to put
>> them online, they become blurred !!
Lee,
Post a cropped sample of an original size and a reduced sample, and the
method you used. Maybe someone can do better (or not). In photoshop, the
bicubic sharper option in the pull-down ought to be a tad sharper.
--
Paul Furman
www.edgehill.net
www.baynatives.com
all google groups messages filtered due to spam
pg
> > <hsyq...@gmail.com> wrote
>
> >> Hello !
>
> >> I do computer graphics as a hobby, and have produced quite a number of
> >> stunning graphics. Often time though, when I shrink the graphic to put
> >> them online, they become blurred !!
>
> Lee,
> Post a cropped sample of an original size and a reduced sample, and the
> method you used. Maybe someone can do better (or not). In photoshop, the
> bicubic sharper option in the pull-down ought to be a tad sharper.
The guy already did that ...
A. www.PenangA1.com/png/1K.PNG
Harris
news:f51b3cf5-3b71-471d...@b9g2000prh.googlegroups.com:
From what I can see, the 1K version is shrinked to smaller width and, as a result, there is an expected
blurring due to pixel aliasing (partial overlaying). I can't make out what the ORI picture shows, but it
seems it's just a larger version of an aliased image section. There's nothing wrong with your software,
these are typical effects in image processing.
--
Harris