Main Page | Report this Page
 
   
Science Forum Index  »  Image Processing Forum  »  Shrinking a picture --- and it becomes blur !!
Page 2 of 3    Goto page Previous  1, 2, 3  Next
Author Message
Marco Al
Posted: Tue Apr 29, 2008 2:37 pm
Guest
bugbear wrote:

Quote:
Since the downsize is an exact integer (factor of Cool
I'm not sure Lanczos (or anything "better" than bilinear)
would help.

Bilinear interpolation is not the same as a box filter at greater than
2x reduction. Using bilinear interpolation at those kind of ratios
wouldn't be very wise.

As for it being an exact integer ratio, that doesn't automatically make
a box filter the best or best looking filter. Pixels are not little
squares (and not little sincs either). Sampling pixels with a box shaped
footprint will create stair stepping.

Marco
KatWoman
Posted: Tue Apr 29, 2008 4:37 pm
Guest
<hsyq8xg@gmail.com> wrote in message
news:52088436-f3ae-42a8-862f-cb283c8c517c@q24g2000prf.googlegroups.com...
Quote:
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 !!

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

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
Posted: Tue Apr 29, 2008 4:50 pm
Guest
bugbear <bugbear@trim_papermule.co.uk_trim> writes:

Quote:
Or try the "Lanczos" algorithm

Since the downsize is an exact integer (factor of Cool
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
Posted: Wed Apr 30, 2008 2:03 am
Guest
In article
<52088436-f3ae-42a8-862f-cb283c8c517c@q24g2000prf.googlegroups.com>,
hsyq8xg@gmail.com wrote:

Quote:
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
Posted: Wed Apr 30, 2008 3:46 am
Guest
tacit wrote:
Quote:
In article
52088436-f3ae-42a8-862f-cb283c8c517c@q24g2000prf.googlegroups.com>,
hsyq8xg@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:

Quote:
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
Posted: Wed Apr 30, 2008 6:09 am
Guest
In rec.photo.digital mark.thomas.7@gmail.com wrote:
Quote:
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 Cool
I'm not sure Lanczos (or anything "better" than bilinear)
would help.

? BugBear

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 cam@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
Posted: Wed Apr 30, 2008 6:55 am
Guest
If you are creating graphics from scratch, that is not from a camera of any
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
Posted: Wed Apr 30, 2008 6:57 am
Guest
"aruzinsky" <aruzinsky@general-cathexis.com> wrote

Quote:
Instead of excess verbosity, you should post links to crops of before
and after images showing the problem areas.

ZING!
Edward Rosten
Posted: Wed Apr 30, 2008 10:04 am
Guest
On Apr 29, 1:37 pm, Marco Al <m.f...@student.utwente.nl> wrote:

Quote:
As for it being an exact integer ratio, that doesn't automatically make
a box filter the best or best looking filter. Pixels are not little
squares

That's exactly what pixels are, in every camera I've ever heard of.
The light is projected on to the focal plane at "infinite" resolution
(ignoring the lense, arpeture, etc), and the CCD integrates the amount
of light falling on a bunch of little squares. IOW, the CCD is
precisely a box filter.

At exact integer ratios is pretty much exactly what using the same
sized CCD chip with fewer (ie larger) pixels would do.

Of course, this isn't a real image, but box filtering is quite
physically accurate for integer ratios or large non-integer ratios.

It can produce stair-stepping, but then again, so can a real camera.

-Ed

--
(You can't go wrong with psycho-rats.)(http://mi.eng.cam.ac.uk/~er258)

/d{def}def/f{/Times s selectfont}d/s{11}d/r{roll}d f 2/m{moveto}d -1
r 230 350 m 0 1 179{ 1 index show 88 rotate 4 mul 0 rmoveto}for/s 12
d f pop 235 420 translate 0 0 moveto 1 2 scale show showpage
Beryl
Posted: Wed Apr 30, 2008 1:29 pm
Guest
N wrote:
Quote:
hsyq8xg@gmail.com> wrote:

....

Quote:
1. Can you tell me of the best way to shrink a 8192X6144 size graphic
to
1024X768 size graphic without losing the interesting details?

....

Quote:
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
Marco Al
Posted: Wed Apr 30, 2008 3:51 pm
Guest
Edward Rosten wrote:

Quote:
That's exactly what pixels are, in every camera I've ever heard of.
The light is projected on to the focal plane at "infinite" resolution
(ignoring the lense, arpeture, etc), and the CCD integrates the amount
of light falling on a bunch of little squares. IOW, the CCD is
precisely a box filter.

The corners deliver less signal, the lens has spherical aberration and
with bayer pattern CCDs you will usually have a diffuser in the optical
path as well. As a system the pixel from most cameras won't have an
image plane footprint very close to a box.

Quote:
Of course, this isn't a real image, but box filtering is quite
physically accurate for integer ratios or large non-integer ratios.

It's mostly physically accurate if your aim is to reproduce the same
effect as that of a Foveon camera without anti-aliasing filter.

Quote:
It can produce stair-stepping, but then again, so can a real camera.

I'd call that a bad camera, not something to strive to reproduce in your
algorithms.

Marco
Allen
Posted: Wed Apr 30, 2008 3:59 pm
Guest
Beryl wrote:
Quote:
N wrote:
hsyq8xg@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
One time when I was a child my mother sent me to the store to buy ten

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
Posted: Wed Apr 30, 2008 4:25 pm
Guest
On Wed, 30 Apr 2008 15:59:52 -0500, Allen <allen@nothere.net> wrote:

Quote:
Beryl wrote:
N wrote:
hsyq8xg@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
One time when I was a child my mother sent me to the store to buy ten
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


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.)
Guest
Posted: Thu May 01, 2008 5:12 am
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

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 ??





aruzinsky wrote:
Quote:
On Apr 29, 12:39�am, 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 !!

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

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."
pg
Posted: Thu May 01, 2008 5:24 am
Guest
On May 1, 8:12 am, hsyq...@gmail.com wrote:
Quote:
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


Quote:
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 ??

aruzinsky wrote:
On Apr 29, 12:39�am, 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 !!

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

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."
 
Page 2 of 3    Goto page Previous  1, 2, 3  Next   All times are GMT - 5 Hours
The time now is Thu Jul 24, 2008 5:47 pm