Question on image format and social media

[pschlute]

I just took a file in LR and exported it twice to my desktop, once set at 240 dpi and once at 100 dpi.

You would have used PPI. DPI is a property of a printer, not an image file. Either way, an image saved at 240ppi or 100ppi will look identical on any screen.

When preparing an image for printing PPI/Pixel dimensions/image size are three corners of the same triangle. If I have an image 2700 pixels x 2700 pixels and I want to print it at 9 inches x 9 inches, it must have a PPI of 300. If you want to use a different PPI you will be either changing the output image size in inches, or up/down sampling the image (changing pixel dimensions) or possibly both.

[Manfred M]

For the record, I made no comment on ppi versus dpi. That was your distinction, not mine.

You said:

In other words, ppi is for printing, not screen viewing.

I suggested that dpi, not ppi are used for printing

[xpatUSA]

You said:

I suggested that dpi, not ppi are used for printing

Pardon my lack of printing expertise. May I ask then: what is the actual purpose of the ppi tags in the EXIF? What are they used for and what changes if their values are made different? Apparently not the size of a print ...

... and yet I thought that a 1500x1200px image printed with those tags set to 150 and "Inches" would actually come out as 10"x8", assuming no scaling interference by the app or the driver.

Help me somebody, please ...

I was actually aware how printers work by squirting splodges of ink onto a medium, just as a pixel on a monitor is made of three RGB sub-pixels.

[pschlute]

... and yet I thought that a 1500x1200px image printed with those tags set to 150 and "Inches" would actually come out as 10"x8", assuming no scaling interference by the app or the driver.

It will come out as a 10x8 print. 1500/150 x1200/150. The purpose of ppi is to achieve resolution (for a resultant print) and define a print size. It is a property of an image file.

The number of dots (as in dpi) that the printer uses to print the image will also affect resolution, but has no bearing on print size. dpi cannot be a property of an image file.

[dje]

No, strictly speaking quality settings are not part of the JPEG standard at all. What they mean in a particular app is made up by the person who added it to that app. Strictly speaking, the quant tables are not part of the standard, either. JPEG included a sample quant table that has ended up being incorporated in all but specialty applications, although that was not the original intent. In some medical applications, researchers have devised more appropriate application-specific quant tables, as was the original intent of the JPEG group. I myself have devised some mods to the example tables in my pre-retirement incarnation as a medical device software engineer. FWIW

ETA: While speaking strictly, I should mention that JPEG is not a file format at all: it is a stream format. What we think of as a JPEG file is generally a JFIF file, which was a very minimal extension of the JPEG definition to support files.

Well perhaps my use of the term “strictly speaking” was not the best choice of words. Note also that I didn’t make any reference to the jpeg standard.

Is it not true that commonly in editing software, when you export a jpeg, there is a setting that affects how the quantisation table is set up and another that selects the chroma sub-sampling to be used? And the user may or may not get to choose these settings individually? And both effect quality?

[Manfred M]

The number of dots (as in dpi) that the printer uses to print the image will also affect resolution, but has no bearing on print size. dpi cannot be a property of an image file.

Agreed. A printer has a single native resolution which is based on the actual print head design; most printers output at 300 dpi; Epson printers (which use a different technology) are higher resolution at 360 dpi. "Best Practice" when preparing an image for print is to rescale it when the image resolution in ppi is the same as the printer's native resolution in dpi. That results in the print maker determining the image quality rather than leaving the scaling to the printer driver.

Computer screens also have a native resolution, but in most cases, that is not most people get too concerned about.

[tclune]

Is it not true that commonly in editing software, when you export a jpeg, there is a setting that affects how the quantisation table is set up and another that selects the chroma sub-sampling to be used?

As I tried to suggest in my first post, what "quality factor" modifies is unknown in general. It attempts to trade off file size and image quality, but how it does that is not part of any public specification that I know about. I presume that different programmers make different decisions on that. It is reasonable speculation that they are modifying or replacing one or both of the quant tables, but whether there is any theoretical justification for what they are doing will presumably vary with the programmer.
Selecting chroma subsampling for getting smaller files is almost assuredly a bad idea. Imaging devices have an actual chroma subsampling, and the choice of which you use in your JPEG image should track that (Bayer is 4:2:2 if memory serves and Fovean 4:4:4, for example.) The actual compression achieved beyond the default chroma quant table by using the wrong subsampling probably will be slight and the impact on IQ potentially significant. So I would expect (although I have not researched this -- caveat emptor!) that the only thing you would want to use the chroma subsampling option for is to align it with the native subsampling of your device. FWIW

[DanK]

Is it not true that commonly in editing software, when you export a jpeg, there is a setting that affects how the quantisation table is set up and another that selects the chroma sub-sampling to be used? And the user may or may not get to choose these settings individually? And both effect quality?

Someone can correct me if I'm wrong, but in the case of Adobe software, I believe the answer is "no". There is a single "quality" setting, and I haven't been able to find anything that explains how it manages those two variables.

If you read the Friedl blog posting I posted, you'll see that he found that the first two steps down on the Lightroom 13-point composite "quality" scale, which he suspects is similar to the Photoshop scale, have essentially no effect on images viewed on screen. that's enough for me, given the other, much more consequential variables entailed in viewing on screen that Bill pointed out.

[xpatUSA]

A printer has a single native resolution which is based on the actual print head design; most printers output at 300 dpi; Epson printers (which use a different technology) are higher resolution at 360 dpi.

I looked up the default ppi for my Sigmas. It is 180 ppi, implying a relationship between them and Epson! That default value doesn't seem to change between models, even though the model's pixel count varies considerably - not to mention their 2x2 true on-sensor 'binning' option.

So the exif tag values as saved are just nominal, since my 2640x1760px wide images at 180 ppi would be a tad too big for my 8x10" Canon printer ...

"Best Practice" when preparing an image for print is to re-scale it when the image resolution in ppi is the same as the printer's native resolution in dpi.

I don't fully understand that. Is the "image resolution in ppi" the value of the exif tags or should we actually re-sample the image to different pixel dimensions?

In other words, would I re-size the image so that it would print at the desired size assuming that the tags have also been set to match my Canon's native resolution?

Example: I want to print a 2640x1760px at 10x8".

I crop the image to 5:4 aspect ratio. I re-sample it to 3000x2400px. I set the image ppi to 300. Voila?

Again, please pardon my ignorance ...

[xpatUSA]

Is it not true that commonly in editing software, when you export a jpeg, there is a setting that affects how the quantisation table is set up and another that selects the chroma sub-sampling to be used?

Agreed, so long as 'commonly' doesn't refer to the size of the Elephant in the Room ...

And the user may or may not get to choose these settings individually? And both affect quality?

Agreed again. I find personally that the chroma sub-sampling selection has a lesser effect on "quality" than the degree-of-compression selection. I habitually post to the web at 4:2:0 and minimum compression. Not saying anyone else should.

Slightly off-topic but in FastStone Viewer, one can select no transformation to CbCr at all, just RGB ...



Not to mention a few other photometrics such as CMYK ...

[dem]

I suggested that dpi, not ppi are used for printing

That would be incorrect. The image does not have dots, it has pixels. By setting 180 pixels to inch in the JPEG metadata one says "when it comes to printing, please fit 180 pixels of this image for every inch of paper". At this point in time we don't know if this image is going to be printed on a printer with 300 dpi, 360 dpi, 600 dpi or 5760 dpi. The dpi is a hardware characteristic of the printer, not a property of a JPEG file.

By matching the ppi of the image to the dpi of the printer, you make life easier for printer's driver and get more control over the final result.

[xpatUSA]

A printer has a single native resolution which is based on the actual print head design; most printers output at 300 dpi

Looks like I'm way ahead of "most printers" - my Canon all-in-one prints at a scorching 9600 dpi!

Or ... "dpi" has more than one correct meaning, as can often occur in the world of photography ...

[dje]

As I tried to suggest in my first post, what "quality factor" modifies is unknown in general. It attempts to trade off file size and image quality, but how it does that is not part of any public specification that I know about. I presume that different programmers make different decisions on that. It is reasonable speculation that they are modifying or replacing one or both of the quant tables, but whether there is any theoretical justification for what they are doing will presumably vary with the programmer.
Selecting chroma subsampling for getting smaller files is almost assuredly a bad idea. Imaging devices have an actual chroma subsampling, and the choice of which you use in your JPEG image should track that (Bayer is 4:2:2 if memory serves and Fovean 4:4:4, for example.) The actual compression achieved beyond the default chroma quant table by using the wrong subsampling probably will be slight and the impact on IQ potentially significant. So I would expect (although I have not researched this -- caveat emptor!) that the only thing you would want to use the chroma subsampling option for is to align it with the native subsampling of your device. FWIW

Here's the luminance quantisation tables and the chroma sub-sampling information taken from four test jpegs, created with Adobe CS6 and data read using jpeg snoop software. The quality settings were Max, High, Med, Low.

Qual Max
Destination ID=0 (Luminance)
DQT, Row #0: 2 2 3 4 5 6 8 11
DQT, Row #1: 2 2 2 4 5 7 9 11
DQT, Row #2: 3 2 3 5 7 9 11 12
DQT, Row #3: 4 4 5 7 9 11 12 12
DQT, Row #4: 5 5 7 9 11 12 12 12
DQT, Row #5: 6 7 9 11 12 12 12 12
DQT, Row #6: 8 9 11 12 12 12 12 12
DQT, Row #7: 11 11 12 12 12 12 12 12
Chroma subsampling: 1x1 (4:4:4)

Qual High
Destination ID=0 (Luminance)
DQT, Row #0: 6 4 7 11 14 17 22 17
DQT, Row #1: 4 5 6 10 14 19 12 12
DQT, Row #2: 7 6 8 14 19 12 12 12
DQT, Row #3: 11 10 14 19 12 12 12 12
DQT, Row #4: 14 14 19 12 12 12 12 12
DQT, Row #5: 17 19 12 12 12 12 12 12
DQT, Row #6: 22 12 12 12 12 12 12 12
DQT, Row #7: 17 12 12 12 12 12 12 12
Chroma subsampling: 1x1 (4:4:4)

Qual Med
Destination ID=0 (Luminance)
DQT, Row #0: 12 8 13 21 26 32 34 17
DQT, Row #1: 8 9 12 20 27 23 12 12
DQT, Row #2: 13 12 16 26 23 12 12 12
DQT, Row #3: 21 20 26 23 12 12 12 12
DQT, Row #4: 26 27 23 12 12 12 12 12
DQT, Row #5: 32 23 12 12 12 12 12 12
DQT, Row #6: 34 12 12 12 12 12 12 12
DQT, Row #7: 17 12 12 12 12 12 12 12
Chroma subsampling: 2x2 (4:2:0)

Qual Low
Destination ID=0 (Luminance)
DQT, Row #0: 18 14 22 35 44 39 34 17
DQT, Row #1: 14 16 21 34 28 23 12 12
DQT, Row #2: 22 21 27 28 23 12 12 12
DQT, Row #3: 35 34 28 23 12 12 12 12
DQT, Row #4: 44 28 23 12 12 12 12 12
DQT, Row #5: 39 23 12 12 12 12 12 12
DQT, Row #6: 34 12 12 12 12 12 12 12
DQT, Row #7: 17 12 12 12 12 12 12 12
Chroma subsampling: 2x2 (4:2:0)

Dave

[xpatUSA]

Here's the luminance quantisation tables and the chroma sub-sampling information taken from four test jpegs, created with Adobe CS6 and data read using jpeg snoop software.

The tables are beyond me but I do have the Snoop too, Dave. Great minds ...

About half-way down, the Snoop also lists a compression-ratio and the bits-per-pixel metric. No idea how they calculate them and sometimes the values are nonsense, i.e. different to mine.

[pschlute]

I don't fully understand that. Is the "image resolution in ppi" the value of the exif tags or should we actually re-sample the image to different pixel dimensions?

In other words, would I re-size the image so that it would print at the desired size assuming that the tags have also been set to match my Canon's native resolution?

If you want 300ppi, then yes you resample the image. But re sampling does not necessarily change the ppi. Have a look at my post above where I described pixel dimensions/ppi/print size as three corners of the same triangle. It is possible to re sample and change the print size leaving the ppi unchanged. When you do re sample make sure you are specifying either a ppi or a document size in addition to the pixel dimensions.

Example: I want to print a 2640x1760px at 10x8".

I crop the image to 5:4 aspect ratio. I re-sample it to 3000x2400px. I set the image ppi to 300. Voila?

Correct. I do this when when I send a file off to a commercial printer who requires files at 300ppi. Although I am usually down-sizing the pixel dimensions of the image.

If one is actually up-sizing an image, there will come a point where the re-sample is downgrading the resolution. One area where more mega-pixels to start with is a very good thing.

[dje]

Someone can correct me if I'm wrong, but in the case of Adobe software, I believe the answer is "no". There is a single "quality" setting, and I haven't been able to find anything that explains how it manages those two variables.

Dan if you look at the data I posted in post 33, you will see that for Max and High Quality settings, chroma sub-sampling is 4:4:4 and for the other two quality settings it is 4:2:0. With the quantisation tables, the values in the top left corner increase as the quality setting goes towards Low. This is an indication of more compresssion (and lower quality).

Dave

[xpatUSA]

If you want 300ppi, then yes you resample the image. But re sampling does not necessarily change the ppi.

Agreed, thanks.

Have a look at my post above where I described pixel dimensions/ppi/print size as three corners of the same triangle. It is possible to re sample and change the print size leaving the ppi unchanged. When you do re sample make sure you are specifying either a ppi or a document size in addition to the pixel dimensions.

Would that be the printed image size or the paper size?

If one is actually up-sizing an image, there will come a point where the re-sample is downgrading the resolution.

Quite so. I use the Mitchell algorithm for up-sizing but I read that dedicated upsizers e.g. GigaPixel are getting pretty good ...

One area where more mega-pixels to start with is a very good thing.

Agreed but I don't print. Academic interest only ...

[xpatUSA]

Looks like I'm way ahead of "most printers" - my Canon all-in-one prints at a scorching 9600 dpi ***

Or ... "dpi" has more than one correct meaning, as can often occur in the world of photography ... LOL

*** Where "d" stands for droplets, not "dots" whatever they are!

I just now researched ppi versus dpi ... the degree of misinformation on the web is quite staggering!

[Manfred M]

If one is actually up-sizing an image, there will come a point where the re-sample is downgrading the resolution. One area where more mega-pixels to start with is a very good thing.

Not just more mega-pixels, but also a larger sensor make a big difference. I find that prints from my full-frame sensor are better than images shot with an APS-C or mFT sensor camera. If I were getting into digital photography today, I would probably invest in a medium format Fuji or Hasselblad camera.

[pschlute]

Would that be the printed image size or the paper size?

It will be the printed image size. Where you are printing without a border, the two will be the same.

I regularly send images off to a printing firm to print at 24" x 16" which I hang in a frame at home. The printer wants the images correctly sized at 300ppi. So I downsize my images to 7,200 x 4,800 pixels at 300ppi. (7,200/300=24" etc)