Jump to content

Recommended Posts

I studied those algorithms back in the dark ages

 

I once wrote a travesty program based on parts of a conditional Huffman encoder. I can't quite recall what language I was using then but it must have been one of the scripting languages in IBM's CMS, such as EXEC or REXX. That must have been in 1984 or thereabout as I had read about the Travesty thing in Scientific American shortly before. I am quite sure I have not done it in Assembly; I had stopped using that language quite some time before.

Link to post
Share on other sites

I once wrote a travesty program based on parts of a conditional Huffman encoder. I can't quite recall what language I was using .

 

 

:) I am guessing it was a an IBM proprietary script. If you wrote in C, then if you are as removed as I am, we cannot read our own code. C is the language we wrote but cannot read.

 

Seriously. I wrote a one-line of code that sent my supervisor to ecstasy for its elegance and fifteen years later I could not understand it.

Link to post
Share on other sites

Advertisement (gone after registration)

Interesting theory.

Not just interesting; it also agrees with the observations.

 

But noise on highlights should be minimal and most of it removed by the binning operation.

Binning? I am not aware of any binning performed.

 

Anyway, while noise in the highlights is minimal (relative to the signal, though not in absolute terms), the highlights are the last area where you would expect to run into posterisation issues. Once you map the linearly digitised raw data into the logarithmical scales of exposure values or f-stops, you get a ridiculously fine resolution of tonal values in the highlights that drops sharply if you turn to the shadows. The lossy sqrt-compression reduces this imbalance but the resolution of highlights is still considerably higher than that of the shadows even after compression is applied.

Link to post
Share on other sites

Binning? I am not aware of any binning performed.

 

It's binning on values, not on pixels. Call it quantization if you like.

 

Anyway, while noise in the highlights is minimal (relative to the signal, though not in absolute terms), the highlights are the last area where you would expect to run into posterisation issues. Once you map the linearly digitised raw data into the logarithmical scales of exposure values or f-stops, you get a ridiculously fine resolution of tonal values in the highlights that drops sharply if you turn to the shadows. The lossy sqrt-compression reduces this imbalance but the resolution of highlights is still considerably higher than that of the shadows even after compression is applied.

 

Not if you postprocess images. Algorithm such as tone-mapping will compress highlights and may cause posterization. Now, if only someone could give me two identical files from an M9, one compressed and one not, I might be able to demonstrate it.

Link to post
Share on other sites

Not if you postprocess images. Algorithm such as tone-mapping will compress highlights and may cause posterization.

The gamma correction will always be applied to provide natural looking tonal values; any postprocessing is applied on top of that. Of course you could tweak the histogram until there is posterisation, but then you could achieve that result even with an uncompressed file.

 

Ever since 2006 there have been suspicions that the lossy compression in the M8 (and later the M9) would create posterisation issues in postprocessing, still these issues have never been successfully demonstrated to exist.

Link to post
Share on other sites

The gamma correction will always be applied to provide natural looking tonal values; any postprocessing is applied on top of that.

 

Not all algorithms are applied after gamma correction. And even so, it really depend on what the algorithm does.

 

still these issues have never been successfully demonstrated to exist.

 

This does not mean they don't exist.

And indeed they exist. As a matter of fact, see CornerFix FAQ (search for "should I use compressed"):

https://sites.google.com/site/cornerfix/faqs

Edited by CheshireCat
Link to post
Share on other sites

Yep, maybe Canon will eventually replace the EOS 7D this photokina, after 5 years.

 

Don't get me started :)

 

In 3 years, Canon launched: 1Dx, 1Dc, 5D3, 6D, 70D, 650D, 700D, 100D, 1200D, M, M2, ...

 

Pro models are usually updated every 2 years (1D) or 3 years (5D).

 

The only notable exception is the 7D (that you picked on purpose :rolleyes:).

However, the 70D, released last year, can be considered a 7D Mark-1.5 (if you don't absolutely need the magnesium body).

Link to post
Share on other sites

Don't get me started :)

 

In 3 years, Canon launched: 1Dx, 1Dc, 5D3, 6D, 70D, 650D, 700D, 100D, 1200D, M, M2, ...

 

Pro models are usually updated every 2 years (1D) or 3 years (5D).

 

The only notable exception is the 7D (that you picked on purpose :rolleyes:).

However, the 70D, released last year, can be considered a 7D Mark-1.5 (if you don't absolutely need the magnesium body).

 

And a Prius is a Maserati without all the extras.

:D

  • Like 2
Link to post
Share on other sites

The Prius can park itself whereas the Maserati can't :D

 

Isn't the 6D some sort of downgraded 5D2?

There is something I find rather annoying about Canon. Leica removed the red dot on the M-P. Doesn't change a thing about my photography.

Canon decided to remove the exchangeable focussing screens on the 5D3. That's annoying as I wanted the better focussing system for my tele lenses and the egs screen for my Zeiss lenses.

I have no problem carrying 2 Leica's for a day, but walking around all day long with a 5D2 with the 21 mm Zeiss plus a 5D3 with a 70-200/2.8???

:confused:

  • Like 1
Link to post
Share on other sites

Create an account or sign in to comment

You need to be a member in order to leave a comment

Create an account

Sign up for a new account in our community. It's easy!

Register a new account

Sign in

Already have an account? Sign in here.

Sign In Now
  • Recently Browsing   0 members

    • No registered users viewing this page.
×
×
  • Create New...