Skip to main content
J
JohnXYZ99_UK
Inspiring
November 17, 2016
Answered

jpeg compression quality not a smooth scale?

  • November 17, 2016
  • 2 replies
  • 377 views

More of an observation than a question. I use Photoshop to edit photos for web display and have noticed that quite often saving at compression value ('quality') 6 produces larger file sizes than for value 7. It appears that the compression algorithm is not following a smooth line, or that perhaps values and 6 and 7 are too close together on the scale? Has anyone else noticed this?

Here are the values (in kB) for a typical landscape photo (1080 x 720 px) with very high detail, from 1 to 10 (6 and 7 highlighted):

200.2, 225.2, 270.0, 300.4, 336.7, 384.6, 377.7, 463.3, 551.2, 692.7

Another photo, moderate detail, 1080 x 1200 px:

168.1, 209.7, 283.5, 326.4, 378.2, 447.0, 434.7, 546.0, 664.0, 842.5

This one was less detailed with plain background (1080 x 900 px):

77.9, 91.3, 108.7, 121.1, 137.7, 164.1, 169.9, 215.6, 268.8, 351.8

As Excel chart:

In the first two, value 6 gives higher file size than value 7, but in the third 7 is higher than 6 as should be the case, though the values in all cases are very close together. I haven't shown values for 11 and 12 - these tend to produce quite a steep line at the end of the scale, with much higher file sizes.

I've experimented briefly with much larger pixel sizes and the same thing seems to happen on shots with high amount of detail, but not with plain backgrounds. Presumably it is simply a peculiarity of the compression algorithm, but it shouldn't really be doing this should it?

John, UK

This topic has been closed for replies.
Correct answer Test Screen Name

As you move up the scale it adjusts a series of numbers. Many of the numbers are quite small. Going from difference 3 to difference 4 can make a big different in quantitisation. There is no 3.5. Also some settings cut the resolution in the H, S channels in 2 or 3 so there's a big jump. Too tech? Don't worry, normal.

2 replies

Mylenium
Mylenium
Legend
November 17, 2016

Presumably it is simply a peculiarity of the compression algorithm, but it shouldn't really be doing this should it?

You would have to ask the development engineers for a proper explanation, but as was already said, the unit values always affect multiple variables of the algorithm, which in reality means that there is never a linear relation to begin with. Even if the individual components would be exposed as separate sliders and you only tweaked one of them it would not necessarily result in linear changes.

Mylenium

T
Test Screen NameCorrect answer
Legend
November 17, 2016

As you move up the scale it adjusts a series of numbers. Many of the numbers are quite small. Going from difference 3 to difference 4 can make a big different in quantitisation. There is no 3.5. Also some settings cut the resolution in the H, S channels in 2 or 3 so there's a big jump. Too tech? Don't worry, normal.

Terms & ConditionsAccessibility statement
Using the community Terms of use Privacy Cookie preferences Do not sell or share my personal information ad choicesAdChoices