Loss of quality when scaling image down.

Hi,

We had some photos taken for our web site, unfortunately the dimensions are
way to big for the web and when the imagea are scaled down the loss of
quality & clarity is very evident. Is there a technique to preserve as much
as possible the quality of an image when scaling it down to a much smaller
size.

Thanks, Jason.

 

Hi Jason,

Welcome to the wonderous land of JPEG (or so I presume).
JPEG is reknowned for its artifacting habit.

When you take pictures, or even save freshly created creations in
Photoshop to a heavy compressed JPEG format image, there's bound to be
ugly artifacting in the picture. That's why it is called a lossy
compression. It throws away detail to get it compressed to smaller
files.

If you have, however, shot the pictures in a less compressed manner
(see the manual of your camera for the settings) you have higher
quality pictures with less artifacting. This will make big files for
your photographs though. It's a weighing decision, have more pictures
with less quality on your cam/memory card, or less photographs but with
a higher quality.

But to get to your problem at hand, those pictures you scaled down...
Are they of bad quality within Photoshop already before you ever saved
them ? And this is with the assumption you do use Photoshop to resize
them. S-Spline Pro is even better for resizing photographs but that's a
commercial add-on on top of Photoshop, so it gets more expensive

Mind you, when resizing within Photoshop, notice the title bar of the
photographs. There's a percentage there. Usually when you open a
larger-than-your-screen-resolution image in photoshop, PS resizes it
realtime to a percentage so that the whole image fits within the
interface. That could have you looking at a preview resizing to say
66%. That looks VERY rough and bad. When you choose the menu
View->Actual size/Pixels then it won't fit fully into the window, but
you are in fact looking at the actual pixels of the picture. When you
have resized, or before resizing choose this, then you're sure you're
looking at the real thing instead of some rough preview-quality
realtime resize of PS.

Hope you have some information to go on. If it's not enough, please reply.

With kind regards,

The PS user.

 

It is also a bit more of a problem than JPEG behavior; downsampling a great
deal can destructive regardless of the final format.

Like the OP, I look forward to some answers, too. I've a number of MF
(6x6cm) and LF(4x5") negative scans to downsample to a good web presentation
and I am very unhappy with the way it has gone.

There might be a math trick - possibly downsampling to a certain frequency
that fits the typical monitor (which I take to be 96 or 120ppi). I've tried
the 10% iteration technique without adequate results.

It seems (and I hope I am wrong) that the negatives have to be rescanned all
over at a modest frequency/resolution.

 

Jason writes ...

Are you talking about high rez film scans or digital capture images?
Should be no problem with digital photos from most 6 - 11 Mpixel
digital cameras I've used ... here are some jpegs of shots my wife took
a few weeks ago in Tanzania with an 8 Mpix camera, we just run USM once
at 300/0.3/0 (per Canon's recommendation), do any tonal corrections
required and then do Image - Image Size with bicubic sharper ...
http://members.aol.com/bhilton665/tanzania_2006/carol_lions_2.htm

With really large film scans sometimes we have reduce in steps but
usually big steps, say 50%, but this is for 6x7 cm scans at 4,000 dpi.

Bill

 

2 writes ...

I took an advanced Photoshop and LightJet printing class from Bill
Atkinson a few years back and he was showing us how he takes 11,000 x
11,000 pixel files (Tango drum scans at 5,000 dpi of 6x6 cm Velvia) and
reducing them to small thumbs, I think 96x96 pixels or so ... he
basically downsampled in 50% steps using bicubic, running a very light
USM step on the 2nd or 3rd iteration on some images (depended on the
data structure). His thumbs looked fine to me, you can find his web
page and see for yourself. This was his method before Photoshop came
out with 'bicubic smoother', which works better at downsampling than
bicubic, so I dunno what he's doing now (though he's switched to
digital so no doubt has a different flow than with film).

Here's an interesting page showing problems with aliasing artifacts
from downsampling by Bart van der Wolf ...
http://www.xs4all.nl/~bvdwolf/main/foto/down_sample/down_sample.htm ...
I downloaded his test pattern and if you downsample it in Photoshop
with one step it does alias but I think (it's been a couple of years)
if you downsample 50% and then downsample a second time to the target
value using 'bicubic sharper' then it looked very good. At any rate
his test pattern is a boundary condition so if you can get it to
downsample smoothly you should be OK with your large format film scans.

Bill

 

Hi there '2',

You should check out, just to make sure, this following website :
http://www.s-spline.com/
They use splines to resize images, which is a hefty mathematical
procedure which imo is pretty darn good. Check out the demo and if you
want to, reply here when you've tested it what you think about the
results it creates ?

With kind regards,

PS user

 

Very helpful! I've got his sampling image, and will do some new trials.

Thanks, Bill (and Atkinson)!

 

I tried it on one of the problematic images: greyscale TIFF, 10301x6250
pixels and the program bombed out, crashed, kaput.

 

when the imagea are scaled down the loss

In my posts I was presuming the problems are caused by downsampling,
not jpeg compression, but I could be wrong. To the original poster,
use 'save for web' and choose the '2 - up' tab to see your original vs
the compressed jpeg version ... I typically try to get 40 - 60 KByte
file sizes (so download times are reasonable for dial-up viewers) with
50% 'quality' setting unless the image has been sharpened too much.
You can see it on the screen so if you have to go with a higher setting
I'd do that rather than settle for poor image quality in the jpegs.
Typically I don't have problems with artifacts or over-compression, I
feel. If the image has been aggressively sharpened then the file sizes
are often much larger.

Bill

 

I wrote ..

My mistake, it's 'bicubic sharper' that works better at downsampling
.... 'smoother' is better at upsampling ...

At least I got it right here

Bill

 

LOL, no, that's not what you want from a commercial tool

 

Yes, I did notice it was ' bicubic sharper', I'll try it tomorrow when I
can get my hands on a Mac. I am hopeful.

 

IrfanView will give you great results.
HTH Richard

 

SNIP

You just beat me to posting it ;-)

The risk of potential down-sampling artifacts is usually
underestimated, especially if one needs to sharpen a bit afterwards.
The trick is to remove all detail in the original that exceeds the
resolution potential of the output size.

For those who think the above page is too theoretical, I've also
applied some of the methods to a filmscan, and it confirms the
predictions:
<http://www.xs4all.nl/~bvdwolf/main/foto/down_sample/example1.htm>
Only high quality down-sampling will avoid potential 'disasters',
like grain-aliasing and regular pattern aliasing, and edge
stair-stepping, and...

Bart

 

The unfortunate truth is that when you downsample, as well as upsample,
you will lose detail. This has nothing to do with file format, but has
everything to do with how many pixels are in your image, and how you
arrived at that many pixels.

In the case of web thumbnails, sharpening is the only solution to end
up with a better looking thumbnail. I have had some luck making them
look better by pre-sharpening before I scale them down using unsharp
mask, a very strong sharpening value, and a pixel factor of 3 to 8.
The object of which is to force the edges of objects out so that when
the image is downsampled there is more obvious divisions between
objects in the image. It also helps to expand the tonal contrast using
an auto levels or something similar before reducing. The more levels
you have when you start sharpening the more levels the computer has to
work with when it is averaging the pixels down. Theoretically this
could not be a bad thing.

Keep in mind that when you are resampling, you are increasing or
decreasing the number of pixels in your image by using interpolation.
This means that if you reduced an image to 25% of its original size,
then every 4 pixels will be averaged into 1 pixel. The relative
brightness of all the color channels in each pixel will be averaged and
flattened into one color brightness for each color. This shrinks the
image, and makes it blurry because everything is being averaged. (Of
course, the same thing happens when you enlarge an image.. 400%
enlargement takes an average of a pixel, and based on the eight pixels
that surround it, averages to create 4 pixels from the single pixel you
started with.

Thumbnails are representative images, and so my philosophy is to just
go ahead and sharpen until they give the viewer some idea what the
thumb represents. Other posts point to different sharpening techniques
(multiple scales, bilinear, etc.) and these are the accepted norm when
dealing with this issue. Sometimes it helps to visualize what is going
on when to scale these images, however.

 

I think it should be sharpened after reducing. The only real problem is
sharpening increases contrast so repeated sharpening gives a washed out
look with a more gritty black & white feel, less soft colors.

 

Irfanview is real handy but does not give the best results. I use it for
speed and convenience.

 

I got a batch action recommended in here that does something similar,
several steps, light sharpening for each. I think it does better than
one step and since it's a batch, it's easy to use. I customized it to
suit my sizes and create 640 pixel image and a 100 pixel thumbnail for
each also and it puts them in a reserved folder where I go to retrieve them.

You can download what I've got (will take some editing) here:
<http://www.edgehill.net/1/?SC=go.php&DIR=Misc/photography/raw-batch/web-batch-action>
click on: "paul-resize-for-web.atn"

move up the folders for more explanation & other options...

 

need more info to evaluate:

1 ) what's the original: size, file format and PPI
2) what's the final JPG size you want. 3) how much JPG compression is
there in the original. Open it up in photoshop, Zoom to 500%, Do you
see lots of squarish patterns? Do you see more noise?


Without knowing these 3 concerns, everyone is just crap guessing. But
it could be right. but it's lucky.
If you can't answer these then you'll never really know the problem.

I'll take a real wild guess that they are JPG compressed at HIGH setting.
You'll have to save them as tif's. Go into each RGB channel and filter
out any JPG patterns on the channel.
Reduce the size first then the resolution ( in case they are above 72 PPI)

Apply an unsharp mask only after you have done everything. It's always last.
Also double up the contast with a curves adjustement or even a second
"hard light" layer set to 50% or less.


finally call the client and get better size resolution. if you can get
300PPI , it will go better. Reduce to 72 PP1 after size reduction.
you could even keep it at 120 PPi or 100 PP1 if too blurry


Good Luck

 

try to use Photoshop CS2 "bicubic sharper" algorithm of scaling images down.

Paul Furman napisał(a):