JPEG – Know thine enemy

[byRo]

It’s worth a few minutes to find out how this works. [details]

[Ed_L]

Very interesting! The best part of your tutorials is not the "how", but the "why". Thank you once again.

Ed

[Gary Richardson]

Have to agree with Ed, reading one of your tutorials is always an education. Looking forward to the next one.

[Janet Petty]

You are an excellent teacher. You explain the complicated by breaking it down into bite sized portions even a blockhead can understand. Thank you. Thank you.

Janet

[JustChecking]

i almost missed this little pearl

just, if i might be so bold, i'd correct few (minor) things
- in point 3) you say "Up to this point we can still get back from here exactly to the original image"... this is not exactly true, because many (eg. hardware, and MMX-based coders) don't use floating point arithmetics, but integers, thus (as the coefficients of transformed data are generally floating point numbers) some information is always lost; ie. even if you omit the quantization phase, certain (very small) amount information is always lost
this is the main reason why JPEG2000 was introduced - wavelets not only can achieve higher compression ratio at comparable quality, but there exist many variants of wavelet transform that are integer to integer (daubechies, cohen, schoeder, et al.)

and a nerd note:
- in 5) "throw away all the zeros" - well, you can't throw them away, unless you would save somewhere which of the coefs were zeros; this would take the same amount of space as saving those zeros... this is solved during the encoding phase by encoding zero using less bits than 8 (well, it's much more difficult, but that'd be for another tutorial )...

and extremely nerd note:
- "(*2) The Discrete Cosine Transformation – Fast Fourier’s kid brother" - well, yes and no; DCT has diagonally different spectral properties, because it lacks the sinus part of the expansion...

sorry for all the boredom...

and thanks a lot for the tut!

[byRo]

You're quite right on all counts, I'm sure.
It was just a question of trying not to scare everybody away!!

I did some reading about JPEG2000 and ended up with the opinion that, although can be much better than JPG, they messed it all up with licensing and, for now anyway, nobody really seems to need it.

You can be sure that when I am able to do tutorials again, I will be consulting you on all the nerdy stuff.
Thanks,

Rô

[JustChecking]

Originally Posted by byRo

I did some reading about JPEG2000 and ended up with the opinion that, although can be much better than JPG, they messed it all up with licensing and, for now anyway, nobody really seems to need it.

you're definitely right, the general public isn't at all interested in it... the only real use of it i've seen so far was for transfer of medical data, where the standardized, powerful and looseless format was missing... although, most of faculties and hospitals use their own formats for years now
on the other hand - the licensing thing - eg. one (widely used) variant of JPEG format itself uses arithmetic coding (type of compression algorithm), which is patented, but noone seems to give a cr*p (just like w/ LZW in GIF; this patent fortunately ended recently)...

Originally Posted by byRo

You can be sure that when I am able to do tutorials again, I will be consulting you on all the nerdy stuff.

rather not you don't wanna the tut to end all messed up and unintelligible, do you?

[rari]

Hi byRo, I'm waiting for the second tutorial of jpeg artifacts.

Where is:
"Next up: Cleaning up JPEG’s"

I always shoot in the high quality, but my images are full of jpeg artifacts and i'm very hurry about this. I tried all plugins and the result is a blurred image, worst than the original. I don't know what can i do!
Can you help me, please?

Thanks

[byRo]

Hi , rari,

Describing the problem is a lot easier than solving the problem.

If jpeg was some sort of an overlay on the image then we could (simply?) use some sort of FFT method to separate out the interferance.

But it isn't - instead of being new information added to the image, it actually represents information which has been subtracted. So, by logic, to get back to the original we would have to supply (i.e. invent) the missing information. Not such an easy task!

Often (or always) "less is better", just try to avoid operations that will enhance the artifacts.
The one place where jpeg artifacts are most visible is in diagonal edges. My favourite tool in this case is the Median filter (duplicate the layer, run Noise>Median, mask hide-all and "paint" the offending edges). Just fixing the diagonal edges is often all you really need to do.

Rô
(south of Brazil? Catarinense? Gaucho?)

[rari]

Hi Rô, thank you for your reply.

I'm from Porto Alegre, Rio Grande do Sul and my english is bad,
but I make an effort to write here. This forum is very good.
Nice to meet you.

I'm trying to save some photos that i had shoot when i travelled to Florida. I bought one small camera there, casio ex-z50, but i didn't like because all pictures have jpeg artifacts shooting in hight quality mode. I'm thinking about change this camera for another that shoot in raw/tiff, maybe this other format don't have this problem...i don't know, but i don't want a big camera.
The worst thing is retouch the noise in people's face, the landscapes is easier.
I have photos with artifacts and out of focus too, very hard to retouch!
Maybe if I post one photo here or in another part of the forum?

Thanks