Why differences in jpegs depending on workflow used to create them?

[Redcrown]

Discovered a mystery I can't solve.

Three ways to convert a raw to jpeg.

1. Open in ACR, pass to Photoshop, save as jpeg.
2. Open in ACR, save as jpeg directly from ACR.
3. Target in Bridge, run Image Processor, save as jpeg.

Same colorspace (Adobe98) and same jpeg quality (10) in all three. But the resulting jpeg files are all different. Not visibly different, but statistically different, and slightly different files sizes on disk.

I load all 3 jpegs in Photoshop as layers. Put top one in Difference mode, display Histogram and read StdDev. I get StdDev values between .5 and 1.1 depending on image content.

Seems the jpegs should be identical. Why not? I thought maybe the jpeg compression inside ACR is different than the one inside Photoshop. That might explain why the save from inside ACR is different. But the Image Processor just uses ACR to convert and Photoshop to compress and save. Why would it be different?

Do the same thing with resizing and the differences go up a lot. StdDev values between 1.5 and 2.2 after a resize from 6720 to 1200.

[Bart_van_der_Wolf]

Discovered a mystery I can't solve.

Three ways to convert a raw to jpeg.

1. Open in ACR, pass to Photoshop, save as jpeg.
2. Open in ACR, save as jpeg directly from ACR.
3. Target in Bridge, run Image Processor, save as jpeg.

Same colorspace (Adobe98) and same jpeg quality (10) in all three. But the resulting jpeg files are all different. Not visibly different, but statistically different, and slightly different files sizes on disk.

I load all 3 jpegs in Photoshop as layers. Put top one in Difference mode, display Histogram and read StdDev. I get StdDev values between .5 and 1.1 depending on image content.

Seems the jpegs should be identical. Why not? I thought maybe the jpeg compression inside ACR is different than the one inside Photoshop. That might explain why the save from inside ACR is different. But the Image Processor just uses ACR to convert and Photoshop to compress and save. Why would it be different?

Do the same thing with resizing and the differences go up a lot. StdDev values between 1.5 and 2.2 after a resize from 6720 to 1200.

Hi,

Don't know the reason, but probably due to different compression quality, despite having it set to 10 in all cases.

I'd first check the file sizes in MB after compression to JPEG. Are they the same size in kB or MB? Also make sure that the downsampled file dimensions are integer multiples of 8, like your 1200 (= 150 x 8 ), but also make sure that the same resampling algorithms are used (probably not). Resampling will have an effect on detail resolution, and thus on compression results.

There is also a possibility that ACR runs 16-bit/channel, while Photoshop runs 15-bit +1 per channel, and the latter will possibly compress 'better'.

A (Windows) program like JPEGsnoop allows you to have a look at the compression tables inside the JPEG file, if you want to ferret out the actual differences, but that will not tell you where they originated from. I have to assume that the same JPEG libraries are used in all modules, so it's more likely to be something that is resampling related, and/or bit depth. At lower compression quality settings there will also be effects of coarser chroma sub-sampling, that may kick in at different value settings for the different modules.

Just some thoughts.

Cheers,
Bart

[Tim Lookingbill]

Here's items I've read in the past in online discussions on the differences of compression quality between ACR jpegs vs Image Processor vs Photoshop.

They're all slightly different mainly between ACR vs Photoshop. Image Processor... is somewhat of a mystery because it's dialog box opens in Photoshop but doesn't necessarily mean it's using Photoshop's "Save As..." compression algorithms since it's a completely different dialog box invoked within Bridge's Menu Tools though it seems to function as a type of action script with a Bridge GUI, not PS GUI.

That's the best I can come up with to explain the slight data analysis differences you're getting.

And I just remembered Save For Web is different as well so you might check the data analysis from that process. Good grief how many does that make just for converting Raw to jpeg? Talk about redundancy!

[Redcrown]

Thanks for the replies. I had not thought about the 16bit vs 15+1bit variable. It's not important, of course, because the differences are insignificant.  I just hate mysteries.