Jump to content

8 bits versus 16 bits, continuing.


t024484

Recommended Posts

Advertisement (gone after registration)

The job of comparing many possible variations of Jaap’s original 16bit DMR picture, have been finalized.

All 12 variations are available for download.

 

What did I do

 

Starting with the original 16 bits picture, I have also truncated it to 14, 12 and 10 bits, simulating a 14,12 and 10 bits A/D converter.

To reduce the bit length, I used resp. 4*Int(X/4), 16*Int(X/16) and 64*Int(X/64)

 

From these 4 pictures, I have made a SQRT compression / decompression, as used for the Leica M8.

Compression: Y = Int((SQRT(X) + 0.5)

Decompression: Z = Y*Y

I checked it against the decompression table in the M8’s DNG file, and it is 100% the same.

 

I also used a logarithmic compression / decompression, because our eyes have logarithmic sensitivities. I used the Natural Log, or LN.

Compression: Y= Int( 33.8*LN((X+33.8)/33.8) +0.5)

Decompression: Z = Int (((33.8*e^(Y/33.8))-33.8) + 0.5)

This Log is modeled to have a slope of 1 at X=0, like the uncompressed data, and to have a value of 255 at x = 64000, that is where the figure 33.8 comes from.

 

 

What are the Results

 

First quick conclusion based upon the picture that I have:

1) First of all I am completely amazed how much you can change to a picture before it becomes visible.

2) Leica made a wise decision to compress to 8 bits.

3) On my screen, I can see no differences between uncompressed pictures with 16,14 and 12 bits. So 14 bits for the M8 seems more than O.K. to start with.

4) Even with 16 bit samples I can see (very) small differences to the original pictures using the M8 SQRT compression / decompression.

The differences though can be almost neglected. Maybe with more critical pictures ( faces ? ) the differences might be more evident.

5) With the Log compression / decompression I could not find any differences down to 12 bit samples. So this is indeed a better algorithm.

6) As from 10 bits, you can see changes in color, although not at all spectacular. So 12 bits for the A/D converter seems to be the minimum to keep the colors intact. I see no benefit for 16 bit A/D converters.

 

Download Site

Over here, all 12 pictures can be downloaded.

Different variants of compressed DNGs

I am looking forward to read your reactions.

 

Hans

Link to post
Share on other sites

  • Replies 85
  • Created
  • Last Reply

Top Posters In This Topic

Top Posters In This Topic

Posted Images

That has been a lot of effort, Hans, with a balanced conclusion. It should take the sting out of the 8-bits discussion.

I still think the DMR produces marginally better files, which throws me back to my different microlenses theory. I still have to do an exact comparison, but the weather conditions...:(

Link to post
Share on other sites

I'd like to see what differences if any show up on a more difficult image, for example one that was under-exposed, or where the photograph has detailed highlights and shadows at the extreme ends of the histogram.

 

As I see it this test would involve starting with a 16-bit raw file and truncating to 14 or 12 bits before conversion to TIFF or before the compression algorithm.

Link to post
Share on other sites

I'd like to see what differences if any show up on a more difficult image, for example one that was under-exposed, or where the photograph has detailed highlights and shadows at the extreme ends of the histogram.

When you can provide me with such pictures, I would be quite happy to repeat this test. It is now almost a hit of a button.

 

Hans

Link to post
Share on other sites

Advertisement (gone after registration)

I'd like to see what differences if any show up on a more difficult image, for example one that was under-exposed, or where the photograph has detailed highlights and shadows at the extreme ends of the histogram.

 

As I see it this test would involve starting with a 16-bit raw file and truncating to 14 or 12 bits before conversion to TIFF or before the compression algorithm.

 

Are you accusing me of exposing properly, Doug? :p:D

I'll have a look, there must be some duds that I saved...

Link to post
Share on other sites

Hallo Hans!

The job of comparing many possible variations of Jaap’s original 16bit DMR picture, have been finalized.

All 12 variations are available for download.

Thank you, I have downloaded them all and "examined"...

Starting with the original 16 bits picture, I have also truncated it to 14, 12 and 10 bits, simulating a 14,12 and 10 bits A/D converter.

To reduce the bit length, I used resp. 4*Int(X/4), 16*Int(X/16) and 64*Int(X/64)

So, in fact, you have decreased the systems dynamic range. Unfortunally, the original image has a low density distribution (with minor overexposures) so, the cutoff of even 6 bits did not alter so much.

From these 4 pictures, I have made a SQRT compression / decompression, as used for the Leica M8.

I also used a logarithmic compression / decompression, because our eyes have logarithmic sensitivities. I used the Natural Log, or LN.

Compression: Y= Int( 33.8*LN((X+33.8)/33.8) +0.5)

Decompression: Z = Int (((33.8*e^(Y/33.8))-33.8) + 0.5)

This Log is modeled to have a slope of 1 at X=0, like the uncompressed data, and to have a value of 255 at x = 64000, that is where the figure 33.8 comes from.

The main result is, that the compression itself does not alter so much and most noteable not in this picture.

Having 16 dynamic (A/D converter) you can expose with a lot of reserves in the shadows and in the highlight. Even with 14 bit of dynamic, if correct exposed, the image will be well. 12 bit of dynamic is a kind of natural limit and all know digital cameras have this resolution (not more and not less).

10 bits, of course, need proper exposion, either in the shadows (than highlights tend to been lost) or in the highlights (than there is no more "footroom" for the shadows).

2) Leica made a wise decision to compress to 8 bits.
Of course, a genious idea
3) On my screen, I can see no differences between uncompressed pictures with 16,14 and 12 bits. So 14 bits for the M8 seems more than O.K. to start with.
Only expensive monitors can reproduce 3x8 bit colors, no (inkjet) print can do more than 3x6 bit colors.
6) As from 10 bits, you can see changes in color, although not at all spectacular. So 12 bits for the A/D converter seems to be the minimum to keep the colors intact. I see no benefit for 16 bit A/D converters.
The changes in color are mainly based in inaccurate quantization (no rounding) and of course in lost of dynamic in relation to perceptiv visualisation.

14 or better 16 bits of dynamic are the vital base for shadow recovering and also - but less accurat - for highlight recovering.

The 8 bit (color) resolution is the primary key for a "high fidelity" color reproduction. The neccessary dynamic (resolution of A/D converter) is much more a question of the scenery you are capturing.

You should continue your work with variations of the color resolution, by using smaller translation tables. As a result, you will watch a significant step from 7 to 6 bits. In this case one must use a more clever solution by quantisation and error distribution.

If the dynamic is large enough, you can use 5 or even 4 bits only for the translation table length whithout noteable visual effects (monitor and printer).

 

I have done all these in the days I have decided to buy a M8 and I have add all this functionality to my own software were I can show all these effects (lower dynamic, lower color resolution and so on).

Unfortunally, this software is working only under operating system OS/2 (originally IBM, now eComStation). But if you have a running OS/2, you can benefit from it...

Link to post
Share on other sites

Are you accusing me of exposing properly, Doug? :p:D

I'll have a look, there must be some duds that I saved...

 

Jaap,

 

Are you still searching for alternative pictures? I would like to finalize this 18/6 bit investigation, and my guns are still smoking.

 

Hans

Link to post
Share on other sites

It has been my experience that bit depth shortcomings usually become most evident in subtle gradations of even color. Cloudless sky, broad large areas of reflection like automobile sheet metal etc. Lower bit depth images (8bit) will tend to show banding in these gradations, where higher bit depth images do not. I am curious if you had the opportunity to check this with any images that have subtle gradations in them. I am not in the studio for a week or so, so can not send samples for a while.

 

Ken

Link to post
Share on other sites

It has been my experience that bit depth shortcomings usually become most evident in subtle gradations of even color. Cloudless sky, broad large areas of reflection like automobile sheet metal etc. Lower bit depth images (8bit) will tend to show banding in these gradations, where higher bit depth images do not. I am curious if you had the opportunity to check this with any images that have subtle gradations in them. I am not in the studio for a week or so, so can not send samples for a while.

 

Ken

 

This has more to do with working 8-bits files in Photoshop and is the main reason to convert any 8-bit file to 16 bits before postprocessing. It is not the same thing as compression in the camea, however. I must confess to the paranoia of converting DMR files to 32 bits for that reason...:o I believe in giving Photoshop plenty of room to shift the bits.;)

Link to post
Share on other sites

So what can we then blame for the somewhat "lesser" colors of the M8 compared to the DMR?

 

When I first got the M8 the first thing I noticed was that the files could handle less processing than the DMR before they showed artifacts(PS+ACR), I had the feeling that the M8 files was somewhat limited and a bit edgy in a digital-noisy-way, even at ISO160.

When I shoot a color checker I see edgy noise in some of the patches even at ISO160, there's almost none with the DMR and the 5D.

Link to post
Share on other sites

Ok great work there Hans...

 

I can't spot any differences either.

 

What about printing them in an epson printer?

I used the compare function in Lightroom, and switched back and forth between 100% crops of two images, so at exactly the same spot on my screen the picture switched from one to the other within a fraction of a second..

It was always on the sideboard of the wagon that I could see very small differences when the picture had been compressed with the SQRT algorithm.

The differences are best described as just a little bit less sharp, and a little bit noisier, but nothing to worry about.

When compressed with the Log algorithm, there are absolutely no differences visible.

 

I have not printed the pictures, could be another useful step to make.

 

Just for completeness, I am trying to calculate the effective value of the relative error of the compressed/decompressed picture to the original picture, with both SQRT and LOG.

So the answer will be one figure as an efective error in percents to each sample after compressing, decompressing. The smaller this figure, the less is the original picture mutilated.

 

Hans

 

.

Link to post
Share on other sites

Use C4, the M8 files are not handled as well by ACR I find

 

Have you tried the ACR 4.5 profile that was posted here at the weekend? It certainly seemed to improve colour from the M8, and I'd be interested to see how it handled that comparison shot you posted last week of the wing of a red car. I think you'd be pleasantly surprised.

Link to post
Share on other sites

Join the conversation

You can post now and register later. If you have an account, sign in now to post with your account.
Note: Your post will require moderator approval before it will be visible.

Guest
Reply to this topic...

×   Pasted as rich text.   Paste as plain text instead

  Only 75 emoji are allowed.

×   Your link has been automatically embedded.   Display as a link instead

×   Your previous content has been restored.   Clear editor

×   You cannot paste images directly. Upload or insert images from URL.

×
×
  • Create New...