What program is best at JPEG compression?

"HEMI-Powered" wrote in message
...
CSM1 added these comments in the current discussion du jour ...

But, I AM interested in this thread because I DO want to
learn. New knowledge is always being created and to ignore it
is to be both stupid and ignorant.

The best way to handle a lot of photo editing is to convert
the original Jpeg image from the camera (if that is what you
camera produces) to a lossless format such as TIFF. Tiff can
be compressed using LZW compression, which is the same
compression scheme as ZIP files. LZW does not compress as much
as JPEG, but the compression is 100% no loss.

As you probably know, my Rebel XT can do JPEG or RAW or both, but
not TIFF. I understand your point, of course, but about Ivory
Soap Pure percent of the time, I can get there with just one
save. When I cannot, I will save the file as a pspimage file so
as to preserve layers and Alpha channel stuff. I do some of the
former but a LOT of the latter to store various kinds of
selections so I do not have to always do global edits when
working on color, brightness/contrast, noise, etc.

You can convert RAW to Tiff in ZoomBrowser EX that was included with your
Canon Rebel XT.

From the EOS DIGITAL Software Instruction Manual Windows (PDF, 3.98 MB).
The Supported Images are JPEG, RAW, BMP, TIFF and PCD (Kodak Photo CD
images).

But I was talking about converting the Jpeg image in a Photo
editor(Photoshop or The Gimp) to a Tiff image, before doing any edits.

By the way, Tiff does support EXIF info.
Tiff is one of the image formats in the EXIF 2.2 specification. The metadata
part of a EXIF Jpeg are TIF tags.

You load the jpeg, and immediately Save As Tiff.
You then work on the Tiff image only, you leave the jpeg as the master
image.

Tiff LZW supports EXIF also, the only difference is a smaller file in most
cases, however there are cases where the Tiff LZW file is actually larger
than the uncompressed TIFF.


After converting to TIFF, you can edit and save as many times
as you wish, with no loss in picture quality.

After you are done with editing, you can then save a final
image as Jpeg. You should only do very limited edits on a JPEG
file.

Every time you save a JPEG you are re-compressing and losing
information.

I know you're replying to the whole NG, but this doesn't happen
to me, although I do understand the non-linear degradation that
begins at about the 2nd re-edit/re-save cycle.

There are claims of lossless JPEG, but not very many photo
editors are supporting the lossless formats. And they are not
common.

Lossless is supported by so few apps and not at all on the web
and Usenet that it is useless. As you suggest, TIFF is better but
has the disadvantage that EXIF cannot be saved if you do LZW
compression, which makes them pretty large. So, again, when I
need to, I use pspimage because it saves EVERYTHING I am doing,
same as PhotoShop proprietary files do and any other good app can
do.

Jpeg and Tiff image formats have been around for years and are
not likely to go away.

As I understand it, JPEG began with a committee or consortia of
about half mathemticians and half real photographers to develop a
new format that could vastly reduce file size beyond the approx.
50% of an LZW TIFF. I haven't read the spec in a long time, but I
vaguely remember that the definition for JPEG=1 out of a possible
100 was to approximate LZW TIFF. And, since photographers are
only interested in continous tone (to the extent that is true for
24-bit color) photos, and they were NOT interested in line art or
raster text, both of which take an awful beating by JPEG unless
you are extremely conservative in your settings. At least, that
has been my experience and what I have heard/seen others say is
true.

Here is a good list of image file formats and the best use.
http://www.scantips.com/basics09.html

Continue reading the following pages. A lot of good
information in those pages

Thanks muchly!

--
HP, aka Jerry
--
CSM1
http://www.carlmcmillan.com
--

CSM1 added these comments in the current discussion du jour ...

As you probably know, my Rebel XT can do JPEG or RAW or both,
but not TIFF. I understand your point, of course, but about
Ivory Soap Pure percent of the time, I can get there with
just one save. When I cannot, I will save the file as a
pspimage file so as to preserve layers and Alpha channel
stuff. I do some of the former but a LOT of the latter to
store various kinds of selections so I do not have to always
do global edits when working on color, brightness/contrast,
noise, etc.

You can convert RAW to Tiff in ZoomBrowser EX that was
included with your Canon Rebel XT.

Assuming that the best quality of JPEG is Okey, Dokey, I could
convert that to TIFF if I were of a mind to convert it to any
lossless format, which I am not. I don't do RAW yet because
despite more than a year of looking, I have year to find a "RAW
for Dummies" book that can get me to the first step in the
learning process. One time. I am interested in all of this, but
until I hear of a truly better mousetrap, my current workflow
works just fine, with a rare need to save in pspimage,

From the EOS DIGITAL Software Instruction Manual Windows (PDF,
3.98 MB). The Supported Images are JPEG, RAW, BMP, TIFF and
PCD (Kodak Photo CD images).

I also have Raw Shooter Premium, but do not have the need yet,
considering the above - don't know how to get to the starting
line for RAW, and don't really want to shoot at 8 mega pixels
when I save to about 1.5 MP eventually.

But I was talking about converting the Jpeg image in a Photo
editor(Photoshop or The Gimp) to a Tiff image, before doing
any edits.

By the way, Tiff does support EXIF info.
Tiff is one of the image formats in the EXIF 2.2
specification. The metadata part of a EXIF Jpeg are TIF tags.

I thought that to get EXIT into TIFF, it has to be the
uncompressed kind.

You load the jpeg, and immediately Save As Tiff.
You then work on the Tiff image only, you leave the jpeg as
the master image.

Tiff LZW supports EXIF also, the only difference is a smaller
file in most cases, however there are cases where the Tiff LZW
file is actually larger than the uncompressed TIFF.

I'll look again, maybe I am harboring a misconception. Thanks.

--
HP, aka Jerry

Bill Tuthill added these comments in the current discussion du
jour ...

What's the difference between PSP's 2x1 1x1 1x1 and PSP's 1x2
1x1 1x1 encoding? I used the former, not the latter. PSP
also has a bunch of weird encodings not encountered in other
software.

My experience with PSP 8 some time ago and 9 more recently is that
1x2 creates larger files but higher quality due to less artifacts
of various types, one insidious one looking for all the world like
noise but it isn't. 2x2, OTOH, produces a much smaller file size;
you need to visually judge the difference in quality to see which
you like better. I only use 2x2 when I am desperate for smaller
size and upping the JPEG compression number is also producing
unacceptable defects. It is a big juggling act and some
experimentation may be necessary for you.

--
HP, aka Jerry

HEMI-Powered wrote:

heavily edited, for brevity

I keep in mind that the "P" in JPEG means "Photographer".

heavily edited

Hello, Jerry:

JPEG stands for "Joint Photographic Experts Group" -- just being a
stickler, dog. ;-)


Cordially,
John Turco

John Turco added these comments in the current discussion du
jour ...

HEMI-Powered wrote:

heavily edited, for brevity

I keep in mind that the "P" in JPEG means "Photographer".

heavily edited

Hello, Jerry:

JPEG stands for "Joint Photographic Experts Group" -- just
being a stickler, dog. ;-)


same thing, I said it's been a long time, but P for Photographer is
close enough for P = Photographic to make the point I was making,
OK by you? thank you.

--
HP, aka Jerry

On Jul 27, 2:53 pm, "HEMI-Powered" wrote:
Martin Brown added these comments in the current discussion du
jour ...

Bill, I don't want to hit this at all hard, but I could swear
that during the beta test for PSP 8, Jasc's Chief Scientist,
the same guy, Kris Zaklika, that wrote the math for DCNR,
said that PSP 8 could decide in a "smart" way which block
size to use. I could be

Either way that statement is incredibily funny.

Why is that "incredibly funny", Martin? I get the feeling that
you are calling me a fool.

No. But I am laughing at the claim that PSPro v8 could decide in a
"smart" way which block size to use when it had one of the most broken
JPEG implementations ever seen in a commercial released product.

PSP 8 has a demonstrably very broken chroma subsampling
implementation in its JPEG codec. It is a testament to just
how robust JPEG is that more people have not noticed this
flaw.

Come again? I do not know either the math or the implementation
but NEVER had any major problems with PSP 8 during the years from
its release until PSP 9 was released,

You were unaware of the problem. So perhaps you don't consider colour
errors a problem. Or you never use chroma subsampling AFAIK the not
subsampled 1x1 1x1 1x1 variant works OK in PSPro 8. NB Photoshop
defaults to not subsampled for the top half of its highest quality
settings and for some images there is a big step in quality there.

The simplest line art test to show its failings on chroma
encoding is to encode a pure red and pure blue 16x16 pokerdot
test pattern in each of the possible phases save as JPEG
default 2x2 subsampled at maximum quality and then reload it.
Half of the newly decoded image not even a vaguely
satisfactory approximation to the original (with gross
residual colour errors - completely the wrong colour bright
red instead of purple). 2x1 subsampling gets an even more
incorrect result with alternating blue and red stripes in
addition.

I don't do line art, don't do vectors, don't do text, don't do
layers, I ONLY do raster graphics bitmap photo/scan editing. So,
pathological examples such as ANY JPEG implementation mangling
line art or text does not surprise me, as that is hardly what
it's authors had in mind. I keep in mind that the "P" in JPEG
means "Photographer". So, I will hardly dispute your conclusions
but do not see how that is relevant to JPEG's real purpose and
usage, other than line art people also want mimimal size images
as well. But, I would think that to maintain maximum quality, you
would want to use a lossless format or even save only to your
favorite editors proprietary format, e.g., pspsimage.

The line art example is constructed to demonstrate a specific
implementation fault in the starkest possible form. However, the
problem is still present in every colour JPEG image made with chroma
subsampling. The damage is less, but there is still unwarranted and
irrecoverable damage to the stored JPEG image coefficients.

Incidentally for some line art JPEG at highest quality not subsampled
can sometimes be the smallest file size. JPEG was designed for
photographic images, but it is quite capable of storing line art with
the right settings (although usually PNG or GIF will be more compact).

For my car pictures, 2x2 subsampling turns them into one big
artifact, I never touch the stuff, so never suffer from its
limitions or damage.

Chroma subsampling tends to hit sharp red or blue colour transitions
disproportionately. And they are rather common colours for cars. But
unless you zoom in to pixel level it should be virtually invisible.
Some of the damage you are seeing could well be from a flawed JPEG
implementation. Check that the problem persists with IrfanView or
GIMP.

Regards,
Martin Brown

On Jul 28, 12:02 am, Bill Tuthill wrote:
In rec.photo.digital Martin Brown wrote:

PSP 8 has a demonstrably very broken chroma subsampling implementation
in its JPEG codec. It is a testament to just how robust JPEG is that
more people have not noticed this flaw.

Thanks for your excellent explanation of the issue.

The simplest line art test to show its failings on chroma encoding is
to encode a pure red and pure blue 16x16 pokerdot test pattern in each
of the possible phases save as JPEG default 2x2 subsampled at maximum
quality and then reload it. Half of the newly decoded image not even a
vaguely satisfactory approximation to the original (with gross
residual colour errors - completely the wrong colour bright red
instead of purple). 2x1 subsampling gets an even more incorrect result
with alternating blue and red stripes in addition.

Is 2x1 same as 4:2:2? I thought the ImageMagick -sampling_factor option
only accepted 1x1 and 2x2, but it does accept 2x1 and produces what looks
(to jpegdump) similar to 4:2:2 chroma subsampling from digital cameras.

2x1 is typically the encoding used by digicams. 1x2 sampling was
unknown until the lossless JPEG transcoder allowed images to be
rotated in the coefficient space. Some decoders don't decode 1x2
chroma especially well.

In principle at least a digicam could encode a Bayer mask sampled
image in a more compact form with 2Y, 1Cr, 1Cb for every 4 sensor
pixels, but the standard had no way of anticipating this development
in all digital imaging sowe are stuck with 4:2:2 where half the pixels
being stored are redundant.

The original is online athttp://www.nezumi.demon.co.uk/photo/jpeg/rb_1x1.jpg
Correct IJG JPEG encodinghttp://www.nezumi.demon.co.uk/photo/jpeg/rb_2x2.jpg

I think you got that filename wrong.

Yes. Should be http://www.nezumi.demon.co.uk/photo/jpeg/rb_2x2_ijg.jpg

And the mess that PSP 8 makes of it at
http://www.nezumi.demon.co.uk/photo/jpeg/rb_2x1.jpg
http://www.nezumi.demon.co.uk/photo/jpeg/rb_2x2.jpg
I would be curious to know if this defect was fixed in later versions.

It appears to be mostly fixed. I'm not certain that the colors are
as accurate as with the equivalent IJG encodings, but they certainly
don't display the wild inaccuracy of the above encodings. I put them
online for you to see:

http://cacreeks.com/psp/ (filenames hopefully self-explanatory)

There seems to be a slight residual systematic error, but it is close
enough now for all practical purposes.

What's the difference between PSP's 2x1 1x1 1x1 and PSP's 1x2 1x1 1x1
encoding? I used the former, not the latter. PSP also has a bunch of
weird encodings not encountered in other software.

One averages the chroma in pairs horizontally and the other
vertically. Empirically it was found that for the standard 4:3 and 3:2
aspect ratios of cameras 2x1 subsampling results in a more compact
representation. The same empiricism explains why the default
quantisation matrix in the standard is not symmetrical.

I wouldn't like to comment on the weird subsampling options offered in
PSPro except to say they may not do what they say on the tin. The ones
in PSPro 8 certainly do not.

Regards,
Martin Brown

Martin Brown added these comments in the current discussion du
jour ...

Martin, I am top-posting only because I wanted to answer
everything easily at one time. For MY work - just my work, not
saying anything about anyone elses - PSP 8 worked fine, and PSP 9
works better. I never had color/chroma problems with 8, don't
with 9. Never had excessive artifacts with either that I didn't
create by blundering to too much compression. Jasc may have made
junk software for many, but it worked fine for me. It is OK that
we disagree, if everyone agreed all the time, a dull world it
would be.

Just for the records, what do you use now? I assume it is NOT PSP
X or XI as I have heard nothing but bad comments about them since
Corel took over.

Have a great week in spite of any friendly controversy. No harm,
no foul.

On Jul 27, 2:53 pm, "HEMI-Powered" wrote:
Martin Brown added these comments in the current discussion
du jour ...

Bill, I don't want to hit this at all hard, but I could
swear that during the beta test for PSP 8, Jasc's Chief
Scientist, the same guy, Kris Zaklika, that wrote the math
for DCNR, said that PSP 8 could decide in a "smart" way
which block size to use. I could be

Either way that statement is incredibily funny.

Why is that "incredibly funny", Martin? I get the feeling
that you are calling me a fool.

No. But I am laughing at the claim that PSPro v8 could decide
in a "smart" way which block size to use when it had one of
the most broken JPEG implementations ever seen in a commercial
released product.

PSP 8 has a demonstrably very broken chroma subsampling
implementation in its JPEG codec. It is a testament to just
how robust JPEG is that more people have not noticed this
flaw.

Come again? I do not know either the math or the
implementation but NEVER had any major problems with PSP 8
during the years from its release until PSP 9 was released,

You were unaware of the problem. So perhaps you don't consider
colour errors a problem. Or you never use chroma subsampling
AFAIK the not subsampled 1x1 1x1 1x1 variant works OK in PSPro
8. NB Photoshop defaults to not subsampled for the top half of
its highest quality settings and for some images there is a
big step in quality there.

The simplest line art test to show its failings on chroma
encoding is to encode a pure red and pure blue 16x16
pokerdot test pattern in each of the possible phases save
as JPEG default 2x2 subsampled at maximum quality and then
reload it. Half of the newly decoded image not even a
vaguely satisfactory approximation to the original (with
gross residual colour errors - completely the wrong colour
bright red instead of purple). 2x1 subsampling gets an even
more incorrect result with alternating blue and red stripes
in addition.

I don't do line art, don't do vectors, don't do text, don't
do layers, I ONLY do raster graphics bitmap photo/scan
editing. So, pathological examples such as ANY JPEG
implementation mangling line art or text does not surprise
me, as that is hardly what it's authors had in mind. I keep
in mind that the "P" in JPEG means "Photographer". So, I will
hardly dispute your conclusions but do not see how that is
relevant to JPEG's real purpose and usage, other than line
art people also want mimimal size images as well. But, I
would think that to maintain maximum quality, you would want
to use a lossless format or even save only to your favorite
editors proprietary format, e.g., pspsimage.

The line art example is constructed to demonstrate a specific
implementation fault in the starkest possible form. However,
the problem is still present in every colour JPEG image made
with chroma subsampling. The damage is less, but there is
still unwarranted and irrecoverable damage to the stored JPEG
image coefficients.

Incidentally for some line art JPEG at highest quality not
subsampled can sometimes be the smallest file size. JPEG was
designed for photographic images, but it is quite capable of
storing line art with the right settings (although usually PNG
or GIF will be more compact).

For my car pictures, 2x2 subsampling turns them into one big
artifact, I never touch the stuff, so never suffer from its
limitions or damage.

Chroma subsampling tends to hit sharp red or blue colour
transitions disproportionately. And they are rather common
colours for cars. But unless you zoom in to pixel level it
should be virtually invisible. Some of the damage you are
seeing could well be from a flawed JPEG implementation. Check
that the problem persists with IrfanView or GIMP.

Regards,
Martin Brown





--
HP, aka Jerry

Martin Brown added these comments in the current discussion du
jour ...

On Jul 28, 12:02 am, Bill Tuthill wrote:
In rec.photo.digital Martin Brown
wrote:

PSP 8 has a demonstrably very broken chroma subsampling
implementation in its JPEG codec. It is a testament to just
how robust JPEG is that more people have not noticed this
flaw.

Thanks for your excellent explanation of the issue.

The simplest line art test to show its failings on chroma
encoding is to encode a pure red and pure blue 16x16
pokerdot test pattern in each of the possible phases save
as JPEG default 2x2 subsampled at maximum quality and then
reload it. Half of the newly decoded image not even a
vaguely satisfactory approximation to the original (with
gross residual colour errors - completely the wrong colour
bright red instead of purple). 2x1 subsampling gets an even
more incorrect result with alternating blue and red stripes
in addition.

Is 2x1 same as 4:2:2? I thought the ImageMagick
-sampling_factor option only accepted 1x1 and 2x2, but it
does accept 2x1 and produces what looks (to jpegdump) similar
to 4:2:2 chroma subsampling from digital cameras.

2x1 is typically the encoding used by digicams. 1x2 sampling
was unknown until the lossless JPEG transcoder allowed images
to be rotated in the coefficient space. Some decoders don't
decode 1x2 chroma especially well.

I can neither confirm nor deny your assertation here, none of my
digitals tell me what they use. I generally do NOT use 2x2
because I find it produces too many artifacts and there is too
much of a tendency for color shifting. My two preferences at 1x2
or 1x1, both of which work fine for my car picturs.

In principle at least a digicam could encode a Bayer mask
sampled image in a more compact form with 2Y, 1Cr, 1Cb for
every 4 sensor pixels, but the standard had no way of
anticipating this development in all digital imaging sowe are
stuck with 4:2:2 where half the pixels being stored are
redundant.

The original is online
athttp://www.nezumi.demon.co.uk/photo/jpeg/rb_1x1.jpg
Correct IJG JPEG
encodinghttp://www.nezumi.demon.co.uk/photo/jpeg/rb_2x2.jpg

I think you got that filename wrong.

Yes. Should be
http://www.nezumi.demon.co.uk/photo/jpeg/rb_2x2_ijg.jpg

And the mess that PSP 8 makes of it at
http://www.nezumi.demon.co.uk/photo/jpeg/rb_2x1.jpg
http://www.nezumi.demon.co.uk/photo/jpeg/rb_2x2.jpg I would
be curious to know if this defect was fixed in later
versions.

It appears to be mostly fixed. I'm not certain that the
colors are as accurate as with the equivalent IJG encodings,
but they certainly don't display the wild inaccuracy of the
above encodings. I put them online for you to see:

http://cacreeks.com/psp/ (filenames hopefully
self-explanatory)

There seems to be a slight residual systematic error, but it
is close enough now for all practical purposes.

What's the difference between PSP's 2x1 1x1 1x1 and PSP's 1x2
1x1 1x1 encoding? I used the former, not the latter. PSP
also has a bunch of weird encodings not encountered in other
software.

One averages the chroma in pairs horizontally and the other
vertically. Empirically it was found that for the standard 4:3
and 3:2 aspect ratios of cameras 2x1 subsampling results in a
more compact representation. The same empiricism explains why
the default quantisation matrix in the standard is not
symmetrical.

I wouldn't like to comment on the weird subsampling options
offered in PSPro except to say they may not do what they say
on the tin. The ones in PSPro 8 certainly do not.

Regards,
Martin Brown


I have learned a lot about the math and science behind these
"strange" number strings. Thank you for taking the time to
educate a long-standing PSP user who is still (apparantly) a
relative newbie.


--
HP, aka Jerry

On Jul 27, 2:45 pm, "HEMI-Powered" wrote:
Martin Brown added these comments in the current discussion du
jour ...

It's not really lossless, because most images are not evenly
divisible by 8 in both directions, so "lossless" rotation
usually trims off some pixels rows or columns on the bottom
or right side.

A good implementation of true lossless rotation refuses point
blank to do it if the requirements for preserving all the
original information are not met. For a crop it is that the
top left corner must start on a block boundary. Unless you are
eagle eyed and have fine detail crossing the edge at an
oblique angle you are unlikely to spot this happening casually
- a lossless crop need not be a multiple of 8 in dimension as
the bottom right corner is unrestricted.

The problem is due to a weakness in the original JPEG spec
which never envisaged that programs would crop or rotate
images in raw coefficient space. It could be almost trivially
fixed by adding a top and left crop offset TAG (2 bytes) to
the existing spec and increasing the size of the image to the
next higher multiple of 8/16.

How big of a problem is lossy compression vs. lossles-but-with-
severe limitations, anyway?

The limitations are no longer all that severe. And they could be
removed entirely if the standards committee decided that it was
worthwhile changing the spec to accomodate recent developments.

Sure, I've mangled a few by over re-
editing my JPEGs when I had no choice, and I've damaged a few
visibly but not catasttrophically, but by doing the re-save
judiciously and by using careful softening of the aliasing and
artifacts that may result, I've always been able to get by.

Oh yes. An image good enough for all practical purposes should be the
target. I am generally interested in finding the smallest image of
adequate quality to be fit for purpose for a given source image.

In this respect wizards that let you compare critical parts of an
image at different compression settings and see the effects in real
time are excellent. And programs that just rename the original image
buffer but leave the original image on display are insidious and
dangerous. Too many people end up overcompressing orginal images by
accident.

don't think I have low standards of quality, but then, I am also
not a purist, theotician, nor extremist.

I suppose I count myself as a pragmatic purist with an interest in the
mathematics.

Given the number of digital cameras about these days I would
hazard a guess that most original source JPEG images are
multiples of 8 in both dimensions (and that a substantial
number are now multiples of 256).

Since my first digital in early 2001, I have immedidately tested
the two or more "quality" or "compression" options available and
always come the conclusion that "standard" or "normal" produces
visible artifacts in a high enough percenteage, maybe 5%, maybe
10%+, I always select "fine", "quality", whichever is the least
compression since I do not use RAW so of necessity I will be
doing at least one edit and one re-save. That philsophy has been
quite successful for me in my type of photography and for my
standards of quality and excellence.

The highest quality in camera JPEG setting is always worthwhile for
maximum flexibility. The only time I ever go down a step has been when
I was in danger of running out of media a long way from any shops. If
it is a choice between slightly inferior images or no images after
running out of media then it is a useful tradeoff in extremis. You
can't do anything similar with film - either you have an unexposed
roll or you don't - there is no option to take twice as many smaller
shots.

Digicams situation is slightly complicated by the default 2x1
chroma subsampling. This means that although lossless rotation
is exact and truly lossless some applications will display a
slightly different image from the rotated coefficients. So for
example:

I have no clue what my Canon Rebel XT uses, but is seems unlikely
that it is 2x1 as I have found that to be a fairly damaging
setting. I don't know, but would think they use either 1x1 (none)
or no more than 1x2.

2x1 is the default for almost all digicams (and it is for the fine
sample rebel XT image on DPreview). The Bayer filter mask makes it non-
sensical to use anything higher. You haven't measured the R or B data
with sufficient spatial resolution to merit anything more. Only Foveon
sensors or scanners sampling RGB at every pixel can justify a 1x1
subsampling.

2x1 should not normally be that damaging unless the source image is
line art with significant pixel wide blue and red stripes. 1x2 was
rarely encountered until the JPEG transcoder allowed images to be
rotated in coefficient space. They are the notation for 2 chroma
pixels average together horizontally or vertically.

Regards,
Martin Brown