Hi Philip,

On 15 Jan 10 10:56 Philip Rhoades said:

- When saving as JPG with 85% quality am I losing information?

Yes!

- How can saving as JPG with 100% quality increase information (file size)?

It doesn't throw so much info away. It's not actually bigger than the the raw data (i.e. total pixels x colour depth) Try saving the image as a BMP to get an indication of that.

- Why is PNG so inefficient?

It's a lossless format (i.e. unlke JPG it doesn't throw any information away). It's not so much "inefficient", rather it just saves ALL the data.

- imageinfo doesn't seem to be able to indicate information structure /compression information differences between the files - is there some way of doing this?

Not in any simple way. You're assumed to know the general benefits of the various file types for different purposes. There is some information in GIMP help. Best you read up some fuller descriptions of the various image file formats. I'm sure someone be be along in a moment with some appropriate links.

Greg

Hi Philip,

Philip Rhoades wrote:

- When saving as JPG with 85% quality am I losing information?

JPG utilizes lossy compression, which means you'll loose information every time you save as JPG, even at 100% quality setting.

That value does not specify the percentage of information stored in the JPG. It is just a number which allows to choose a trade-off between subjective image quality and file size.

In consequence, the workflow recommendation is to routinely save as XCF and only create a JPG when the (finished) work leaves your system.

- How can saving as JPG with 100% quality increase information (file size)?

As said above, it is wrong to assume you were saving 100% of the 85% of the original image's information here.

In practice, one has to look at the compression artifacts to be able to adjust for minimum file size at acceptable quality. There's no way to just rely on the numbers.

What happens in detail:

when opening the JPG it gets decompressed to 2048x1536 RGB pixels of 3 bytes each, a whopping total of 9437184 bytes of RAM. (This holds true for any color JPG of 2048x1536 size, regardless of file size).

Now when saving this image as JPG, it's these 9437184 byte of image that get compressed, regardless from where this data originated.

Compressing an image of 9437184 bytes at 100% gives a larger file size than compressing the same image at 85%. There's no memory of previously used compression rates.

And regardless of file size, each new JPG compression step adds new artifacts to the image, degrading quality.

- Why is PNG so inefficient?

PNG offers lossless compression and isn't designed for use with photos. It excels at graphic data which e.g. has uniform color areas.

regards, peter

Philip Rhoades wrote:

What still doesn't make sense is that if the original file is JPG and one simply opens it and then saves it as another JPG file with 100% quality - you are saying that introduced artifacts are adding about 150% to the file size? (681 KB to 1.618 MB) How could the compression algorithms be so different as to cause this sort of result? - At worst I would have expected maybe a 10% increase in size . .

well firstly, 1.6MB are not that bad in comparison to 9MB of raw RGB data, right? (just try saving to uncompressed BMP as Greg suggested).

Btw, 100% quality for JPG gives very little visible advantage over the default 90% setting.

The relationship between quality value, file size and perceived image quality is very delicate. Any assertion has to be made with a lot of weasel words. So yes, compression artifacts have a tendency to hinder compression, resulting in larger file sizes after re-compression to comparable quality. A similar effect is caused by noise. So to get optimal JPG files, it's best to use a RAW->XCF->JPG workflow where the JPG is created only once.

Just have a look at the advanced settings in the JPG save dialog to get a first impression of what machinery is at work here. There also was a very long thread on gimp.developer on that very quality setting...

How does image quality compare if you adjust the quality slider such that the resulting file size is about 680KB?

regards, peter

On Fri, 15 Jan 2010 21:56:40 +1100, Philip Rhoades wrote:

- Why is PNG so inefficient?

PNG is not efficient for real life images (ordinary photos). PNG is very efficient for computer generated images (like a snaphot of a program window, or a relatively simple paint, or vector graphics, or some CAD drawing export), where large areas use exactly the same pixel value from one pixel to the other.

Cristi

Hi Philip,

On 15 Jan 10 12:53 Philip Rhoades said:

What still doesn't make sense is that if the original file is JPG and one simply opens it and then saves it as another JPG file with 100% quality - you are saying that introduced artifacts are adding about 150% to the file size? (681 KB to 1.618 MB)

The %age, isn't one of straight file size reduction, but relates to the areas of the image to be averaged. The the more you reduce the quality the more you increase the potential for larger areas to be averaged. A "busy" image will not suffer much averaging regardless of the compression requested, but one with large areas of broadly similar colour, sky, painted walls, car bodywork, etc will have progressively larger areas averaged the more you reduce quality and the file size will reduce accordingly.

Some images, for example, where there is a load of tumbling water and spray will barely reduce in size t all even at higher compression levels as no part of the image has a large enough plain area to allow it to be averaged.

How could the compression algorithms be so different as to cause this sort of result? - At worst I would have expected maybe a 10% increase in size . .

I think you are assuming the whole imaged is compressed equally, regardless of the level of detail and colour change from one pixel to the next. Only in images with large areas of similar colour will the fle size reduce much at higher compression levels. The control is more one of "reduce this if you can", rather than "reduce it whatever the consequenses".

Greg Chapman
http://www.gregtutor.plus.com
Helping new users of KompoZer and The GIMP

On Fri, Jan 15, 2010 at 4:56 AM, Philip Rhoades wrote:

People,

I am trying to work out why there is such a large file increase when I edit a file and save it. The background info:

Google the difference between "lossy" and "lossless" image compression. Once you understand the fundamental differences your question will be answered (actually, it will be a non-question) :)

Philip Rhoades writes:

What still doesn't make sense is that if the original file is JPG and one simply opens it and then saves it as another JPG file with 100%

Because JPEG isn't meant to be saved at 100% quality.

The JPEG FAQ, http://www.faqs.org/faqs/jpeg-faq/part1/section-5.html, says:

Except for experimental purposes, never go above about Q 95; using Q 100 will produce a file two or three times as large as Q 95, but of hardly any better quality. Q 100 is a mathematical limit rather than a useful setting. If you see a file made with Q 100, it's a pretty sure sign that the maker didn't know what he/she was doing.

Do a web search on jpeg quality "100%"
and you'll find lots of detailed discussions of this.

GIMP's "Show preview in image window" check box is extremely helpful, and lets you see the trade-off in quality versus size. It's too bad it's not enabled by default.

...Akkana

Hi Philip,

On 15 Jan 10 18:27 Philip Rhoades said:

- when the JPG is uncompressed by GIMP into RAM, there is no loss of information (?)

No further loss, but the restored image is subject to those averages created when the image was originally compressed.

- when GIMP then saves the same image as a new JPG at 100% quality (I would have thought that this meant not losing any more information),

You shouldn't take 100% too literally. Think of it more as "best quality", but the best that the JPG algorithm achieves is not "no change".

that the second JPG would be compressed/created in much the same way as the first and therefore would be about the same size . .

Remember that it is working on data that is already corrupted so it further corrupts it, again averaging the larger areas of similar colour. If you didn't want it to compress it a bit (even at 100% quality) you wouldn't be using selecting a JPG format when saving it would you? You'd choose a lossless format instead.

of course I have no control over the file format that the camera uses

Most cameras do have a range of compression options available within their menu system. It won't have the variability of the GIMP but probably will have a "Normal" setting plus a high and low option.

and cropping a camera image and actually getting a result that is 2.5 times the size of the original is a bit annoying . .

But now you realise how much the data is compressed at even "normal" levels, you realise how good the algorithm was that the JPEG came up with! :-)

Greg Chapman
http://www.gregtutor.plus.com
Helping new users of KompoZer and The GIMP

On Fri, Jan 15, 2010 at 12:27 PM, Philip Rhoades wrote:

- there was a loss of information when the first JPG was saved in the digital camera memory from the CCD

Correct

- when the JPG is uncompressed by GIMP into RAM, there is no loss of information (?)

Since JPG is not lossless, there is always a loss of information. Or more specifically the same JPG can be interpreted differently by different software, so opening it in GIMP might look different than another program perhaps. Once an image is saved as JPG there's no way to get the original image back from that JPG file.

- when GIMP then saves the same image as a new JPG at 100% quality (I would have thought that this meant not losing any more information), that the second JPG would be compressed/created in much the same way as the first and therefore would be about the same size . .

100% quality does not mean no loss of info, just means that it's as close to the original as JPG is capable of getting. It's still not going to be identical to the original.

Good to know that this happens anyhow . . of course I have no control over the file format that the camera uses and cropping a camera image and actually getting a result that is 2.5 times the size of the original is a bit annoying . .

That's why the expensive/professeional cameras output in raw format, so the photographer can have total control. :) FWIW, if you have a Canon you very well might be able to install CHDK and get raw images, that's what I've done with my SD550 and SD1000.

Depending on your purpose for the final JPG file, there are other ways to make the file smaller. Saving as progressive vs baseline usually makes a small difference in size. You can use the program jpegoptim to optimize (losslessly) and reduce filesize, and you can strip out EXIF/thumbnail/etc header info to make the file smaller without affecting the actual image data either (assuming you're not using the EXIF for rotation/etc in whatever program you're displaying the image in).

For best quality/size trade-off when you save your JPG in GIMP I would use the "show preview in window" (something like that) option which will show the resulting compressed JPEG in a window. Then you can adjust the quality and other settings and see how it looks. I usually adjust it as low as I can until the image appears to degrade, then I bump it up a notch and choose that. For me, around 75% seems to be a pretty good combination of small file size and good quality.

There are also JPEG tools that allow you to losslessly crop/rotate etc your pictures. Check out jpegtran from jpegclub.org or one of these programs which supposedly include the same functionality: http://jpegclub.org/losslessapps.html

Hi,

On Sat, 2010-01-16 at 05:27 +1100, Philip Rhoades wrote:

I guess what is confusing is this:

- there was a loss of information when the first JPG was saved in the digital camera memory from the CCD

- when the JPG is uncompressed by GIMP into RAM, there is no loss of information (?)

- when GIMP then saves the same image as a new JPG at 100% quality (I would have thought that this meant not losing any more information), that the second JPG would be compressed/created in much the same way as the first and therefore would be about the same size . .

Actually, you get almost no further degradation if you save the image again with the same settings that were used for the first save. The JPEG plug-in even stores information in the image when the image is opened and it will use that information to save it in the best possible way when you save it again. Just leave all controls at their default values.

Note that I said "almost". Of course the image will suffer a little. But you won't get significantly better results if you increase the JPEG quality or change other settings in the save dialog. You just get a larger file.

Sven

Actually, you get almost no further degradation if you save the image

again with the same settings that were used for the first save. The JPEG plug-in even stores information in the image when the image is opened and it will use that information to save it in the best possible way when you save it again. Just leave all controls at their default values.

Note that I said "almost". Of course the image will suffer a little. But you won't get significantly better results if you increase the JPEG quality or change other settings in the save dialog. You just get a larger file.

Sven

Quality is relative to what you need, how the image is used as well. If it's for internet use, 70% or so is reasonable quality. Image weight adds up fast if you have a lot of large images. I don't do print media so someone else would need to talk to it, but I commonly here that 300-600 dpi is requested whereas for the internet resolution is much less of a factor.

If all I've been given is a .jpg I'll typically save it as a .png along the way.

On Fri, 15 Jan 2010 21:56:40 +1100, Philip Rhoades wrote:

- When saving as JPG with 85% quality am I losing information?

Yes, but still with the same 85% quality you may obtain different results by changing other parameters.

Just look at the following example. Note the file size for each, but most of all, look at the color quality and the outline of objects (the files are quite small; save them somewhere and look at them by switching forth and back so you can notice the differences). The JPEG were both saved with 85%, but one with subsample for best quality and the other with subsample for minimum file size:

http://www.secarica.ro/misc/monopoly.bmp http://www.secarica.ro/misc/monopoly.png http://www.secarica.ro/misc/monopoly_gimp_hi_quality.jpg http://www.secarica.ro/misc/monopoly_gimp_minim_size.jpg

Take the .bmp and do further tests with the save options.

Cristi

On 15.01.2010 19:59, Greg Chapman wrote:

- when GIMP then saves the same image as a new JPG at 100% quality (I would have thought that this meant not losing any more information),

You shouldn't take 100% too literally.

Especially if the value is not a percentage.

Regards, Michael

On Fri, 2010-01-15 at 13:32 -0600, Paul Hartman wrote:

On Fri, Jan 15, 2010 at 12:27 PM, Philip Rhoades wrote:

- when the JPG is uncompressed by GIMP into RAM, there is no loss of information (?)

Since JPG is not lossless, there is always a loss of information. Or more specifically the same JPG can be interpreted differently by different software, so opening it in GIMP might look different than another program perhaps.

I'm fairly sure this is not true - there is only one way to uncompress a JPG file, so all programs should create the same uncompressed version.

-- David

On Sat, 16 Jan 2010 10:54:23 +1100, David Hodson wrote:

I'm fairly sure this is not true - there is only one way to uncompress a JPG file, so all programs should create the same uncompressed version.

Not true - I know that at least different versions of ImageMagick will decompress the same JPEG slightly differently, probably due to different rounding.

On Fri, Jan 15, 2010 at 5:54 PM, David Hodson wrote:

On Fri, 2010-01-15 at 13:32 -0600, Paul Hartman wrote:

On Fri, Jan 15, 2010 at 12:27 PM, Philip Rhoades wrote:

- when the JPG is uncompressed by GIMP into RAM, there is no loss of information (?)

Since JPG is not lossless, there is always a loss of information. Or more specifically the same JPG can be interpreted differently by different software, so opening it in GIMP might look different than another program perhaps.

I'm fairly sure this is not true - there is only one way to uncompress a JPG file, so all programs should create the same uncompressed version.

From an old JPEG FAQ:

"Another important aspect of JPEG is that decoders can trade off decoding speed against image quality, by using fast but inaccurate approximations to the required calculations. Some viewers obtain remarkable speedups in this way. (Encoders can also trade accuracy for speed, but there's usually less reason to make such a sacrifice when writing a file.)"

Also, of course, things like color management and other kinds of post-processing can cause differences when the same file is opened in different programs.

Philip Rhoades wrote:

I am trying to work out why there is such a large file increase when I edit a file and save it. The background info:

Original file (from digital camera) - format, size, depth, geom:

JPEG 680590 8 2048x1536

[..]

After opening, cropping and saving original file as a PNG file with defaults (compression 9):

PNG 4722953 8 2048x1536

There's something funny there. If the image has been cropped, how can it be the same dimensions as the original (2048x1536) (unless you have then done a "resize" you didn't mention)?

On Fri, 2010-01-15 at 18:07 -0600, Paul Hartman wrote:

On Fri, Jan 15, 2010 at 5:54 PM, David Hodson wrote:

On Fri, 2010-01-15 at 13:32 -0600, Paul Hartman wrote:

[...] the same JPG can be interpreted differently by different software, so opening it in GIMP might look different than another program perhaps.

I'm fairly sure this is not true - there is only one way to uncompress a JPG file, so all programs should create the same uncompressed version.

From an old JPEG FAQ:

"Another important aspect of JPEG is that decoders can trade off decoding speed against image quality, by using fast but inaccurate approximations to the required calculations.

Good point. I would argue that's not really interpreting the JPG differently, just being more or less accurate in your interpretation. However, the user may still see some minor differences.

Also, of course, things like color management and other kinds of post-processing can cause differences when the same file is opened in different programs.

Of course.

-- David

On Fri, Jan 15, 2010 at 09:56:40PM +1100, Philip Rhoades wrote:

People,

I found this thread about jpeg veeeeery interesting indeed. For this I thank you all, who asked and who have used his/her spare time to enlight the audience.
I even think that this thread could be posted in a FAQ/wiki/manual since IMHO is a topic that many have often wondered about.

bye

Philip Rhoades wrote:

It still seems counter intuitive that opening a JPG (even if it is a photo rather than a computer generated image) and immediately saving it with 100% "quality" increases the size by 2.5 . .

so you mean the scale should be different? Like

1 .. 10 ... 100 ... 100000 ^ ^
| \ extravagant luxury quality for the filthy rich current "90"
default

i'm not shure if that would not create even more confusion...

regards, peter

PS: as Michael Schumacher previously noted, the quality value is indeed just a number, not a percentage.

Philip Rhoades wrote:

Peter,

On 2010-01-18 20:40, yahvuu wrote:

Philip Rhoades wrote:

It still seems counter intuitive that opening a JPG (even if it is a photo rather than a computer generated image) and immediately saving it with 100% "quality" increases the size by 2.5 . .

so you mean the scale should be different? Like

1 .. 10 ... 100 ... 100000 ^ ^
| \ extravagant luxury quality for the filthy rich current "90"
default

?? - that's an odd comment . .

oh yeah, i should have been more clear. Now that you and me and probably a few others have learned something new about JPG peculiarities, i was brainstorming how the user interface could be tweaked to avoid misleading associations.

Above diagram was intended to depict a logarithmic scale for the quality value, where the numbers relate to the typical growth in file size [1].

But a much better and simpler idea is to just use a number range from 1..13, similar to photoshop.

I'll take that over to the developer's list.

regards, peter

[1] http://en.wikipedia.org/wiki/File:Quality_comparison_jpg_vs_saveforweb.jpg

Philip Rhoades wrote:

Cristi,

On 2010-01-16 06:55, Cristian Secară wrote:

On Fri, 15 Jan 2010 21:56:40 +1100, Philip Rhoades wrote:

- When saving as JPG with 85% quality am I losing information?

Yes, but still with the same 85% quality you may obtain different results by changing other parameters.

Just look at the following example. Note the file size for each, but most of all, look at the color quality and the outline of objects (the files are quite small; save them somewhere and look at them by switching forth and back so you can notice the differences). The JPEG were both saved with 85%, but one with subsample for best quality and the other with subsample for minimum file size:

http://www.secarica.ro/misc/monopoly.bmp http://www.secarica.ro/misc/monopoly.png http://www.secarica.ro/misc/monopoly_gimp_hi_quality.jpg http://www.secarica.ro/misc/monopoly_gimp_minim_size.jpg

Take the .bmp and do further tests with the save options.

Yep, the only one where I could see a difference was with the last one.

It still seems counter intuitive that opening a JPG (even if it is a photo rather than a computer generated image) and immediately saving it with 100% "quality" increases the size by 2.5 .

The exampe image is actually not well-suited for JPG. For this one, indexed PNG would work much better. Som, this comparisonis actually misleading. It would actually be better use a photo for this.

When I was still using Windows (I'm under Ubuntu "Karmic" now), I used Ulead's SmartSaver, where one can view the original & the optimized image at the same time and tweak the settings until one is satisfied with the result.

I usually have compression set to '90%'. Due to quality reasons, I never go below that. Additionally, for DCT (Discrete Cosinus Transformation), I am using 'Fast Integer', not 'Floating Point' (further loss, AFAIK). ('DCT' is part
of the JPG compression process, which works in several steps.)
Claus

< snip >

But a much better and simpler idea is to just use a number range from 1..13, similar to photoshop.

I'll take that over to the developer's list.

Is this any help, I came across it a long time ago?

Here is a table that provides an approximate mapping between Photoshop quality levels and GIMP (actually IJG JPEG library) quality levels:

Adobe Photoshop quality 12

Norman

But a much better and simpler idea is to just use a number range from 1..13, similar to photoshop.

I'll take that over to the developer's list.

I disagree, I think Photoshop's way of displaying the JPG compression slider is ridiculous. You can move the slider back and forth within a very wide range before the corresponding number changes. And since the slider doesn't spring back to a pre-determined spot on the line, that means one could select a different grade of "level 8" depending on where the slider is positioned. You can actually see this by looking at the file size. If I select the lower range of "level 8", the file is smaller than if I pick the higher range of "level 8". This has always bugged me since the dawn of Photoshop (I started using it at version 4.0 back in 1997). I far prefer the Gimp method of displaying this setting.

Is this any help, I came across it a long time ago?

Here is a table that provides an approximate mapping between Photoshop quality levels and GIMP (actually IJG JPEG library) quality levels:

Now that's awesome. I always wondered what the correlation was. And I always worried that Adobe was ignoring the subsampling aspect, now I know better. It's also interesting to see that they don't allow anything lower than 85@2x2.

Frank Gore wrote:

But a much better and simpler idea is to just use a number range from 1..13, similar to photoshop.

I'll take that over to the developer's list.

I disagree, I think Photoshop's way of displaying the JPG compression slider is ridiculous. You can move the slider back and forth within a very wide range before the corresponding number changes. And since the slider doesn't spring back to a pre-determined spot on the line, that means one could select a different grade of "level 8" depending on where the slider is positioned. You can actually see this by looking at the file size. If I select the lower range of "level 8", the file is smaller than if I pick the higher range of "level 8".

I fully agree that this is bad design. If there are interstages of say, 8.0 8.25 8.5 and 8.75, those clearly have to be displayed.

regards, peter

Norman Silverstone wrote:

Here is a table that provides an approximate mapping between Photoshop quality levels and GIMP (actually IJG JPEG library) quality levels:

Adobe Photoshop quality 12

[..]

wow, i grossly underestimated the influence of the advanced parameters: Subsampling 1x1 ist even more important than the quality slider for images with sharp edges..

In the meantime, i stumbled across a blog entry which also compares Photoshop's 'save to web' results:
http://blogs.gnome.org/raphael/2007/10/23/mapping-jpeg-compression-levels-between-adobe-photoshop-and-gimp-24/

regards, peter

yahvuu wrote:

Norman Silverstone wrote:

Here is a table that provides an approximate mapping between Photoshop quality levels and GIMP (actually IJG JPEG library) quality levels:

Adobe Photoshop quality 12

Sure; subsampling takes groups of 4 x 4 pixels and averages the values for hue/lightness for 2 x 2 pixels, or all four pixels respectively.

4 x 4 x 4 retains the original data, 4 x 2 x 2 averages for two groups of 2 pixels each, and 4 x 1 x 1 assigns the same value to all four pixels in the group , which results in a smaller file, but loss of quality at the same time. (That's at least how I understood it.)

4 x 4 x 4: 4 x 2 x 2: 4 x 1 x 1: --------- --------- --------- | | | | | | | | --------- | | | | | | | | | | | | | --------- --------- ---------

Then there' s also the DCT value (Discrete Cosinus Transformation; the algorithm used in JPEG; I think JPG 2000 uses a Discrete Wavelet Transformation instead, which results in better quality -> less JPG artifacts). I have set this option to 'Fast Integer'; 'Floating Point' results
AFAIK in additional loss of information.

Claus

On Mon, 2010-01-18 at 10:40 +0100, yahvuu wrote:

Philip Rhoades wrote:

It still seems counter intuitive that opening a JPG (even if it is a photo rather than a computer generated image) and immediately saving it with 100% "quality" increases the size by 2.5 . .

That is only non-intuitive because you are making the wrong assumption that the JPEG quality was measured in percent. It isn't, it's just a quality level that happens to have a range of 0 to 100. I suggest that you read the JPEG FAQ, in particular the answer to question 5:

http://www.faqs.org/faqs/jpeg-faq/part1/

Sven