Jump to content
Sign in to follow this  
t024484

The effects of compressing to 8 bits

Recommended Posts

Advertisement (gone after registration)

DMR against M8 pictures will still be comparing apples with pears. There are so many differences, like sensor noise, read amplifier noise, A/D converter noise.

Then you have the ISO effect, the higher the ISO value, the more noise.

I did several tests with DMR pictures that where compressed to 8 bits and expanded again.

To my opinion this test comes as close as possible in finding out what 8 bit compression does.

 

On faces I could not find that many differences, but on a macro picture of a yellow Rose, with many subtle color shifts in the middle bright area, I could definitely see differences, but only on a printout, hardly on my LCD.

This outcome corresponds to my graph in posting #4, where the highest deviation of .5 EV is seen in the middle gray tones.

I performed all this out of curiosity. I can only repeat what I have said before, compressing is something we can live with, but it would be nice if Leica gave us the option to decide.

 

Hans

Share this post


Link to post
Share on other sites

One thing I forgot to mention:

I find it most contradictory that Leica turns up every stone upside down, cost no issue, to get their lenses to the highest possible standard, coming with updated lenses one after another.

At the same time they are "throwing away" valuable information by using an 8 bit compression scheme.

That is what strikes me most of all.

 

Hans

Share this post


Link to post
Share on other sites
maybe a little compression noise added is not a bad thing...see here

 

Hi Robert - this is a misunderstanding that was addressed in one of the earlier posts - what's added is not noise, but in fact the opposite of noise: posterization.

 

Ironically enough, the post you point to there actually illustrates perfectly what's wrong with Leica's scheme, which introduces the gradient banding into an image which would otherwise be a smooth tonal gradient.

 

As so many have said before - so strange and sad that the output from the M8 doesn't allow for the full character of the lenses to shine through - and that there should be so much breast-beating about the lack of an AA filter, but so little concern that massive amounts of color information are disposed of before the image is written to the card.

Share this post


Link to post
Share on other sites
Mani, please post an example

 

So my theoretical problems are not good enough for you?

 

Okay - when I get home tonight I'll try to find some example of what I'm talking about...

Share this post


Link to post
Share on other sites

Advertisement (gone after registration)

You know, I agree with you 100% that I want to have the choice. I don't think Leica should take this choice away from us, even if in the end we might not be able to see the difference. But I still have never seen a shot which had posterization caused by the 8-bit scheme, and I would like to see one (crop + raw).

Share this post


Link to post
Share on other sites

I'm another in agreement. Let's have the choice so we can see for ourselves.

 

As for noise, genuinely random noise can be good and when I did digitial audio in the late 80's/early 90's, it was common to add dither (a random number, mean value 0.5 LSB and I tended to use a triangular probability density function) to a processed digital signal (long internal word length) to effectively round off the lowest order bit I wanted to keep before tuncating the fractional digits. That had the beneficial effect of the signal fading into a benign noise floor and it is of course still possible to hear a signal buried in the noise.

 

Here, we have an additional form of noise introduced by the mapping process and the issue is that the noise is not random but related to image content because it is added on to force the data to one of 255 or 256 discrete levels - posterization. In digital audio, such antics make the sound coarse and fatiguing, in digital imaging, we know the familiar banding. Doesn't show up all (much of?) the time of course, and the eyes, like the ears, are quire forgiving.

 

Nikon provide 3 options on the D3/D3x - compressed (lossy), lossless compression, uncompressed. The Leica M8 is currently providing a more lossy compression than Nikon's own with no lossless or uncompressed options. They probably don't have the horsepower to go to lossless compression so uncompressed would be a welcome alternative, say, as an option when you are shooting DNG only.

 

One thing we can be totally certain of is that they won't be doing any such thing in the S2. They didn't in the DMR, so why are M8 owners being penalised?

Share this post


Link to post
Share on other sites
One thing we can be totally certain of is that they won't be doing any such thing in the S2. They didn't in the DMR, so why are M8 owners being penalised?

 

Smaller battery or weaker processing?

Share this post


Link to post
Share on other sites

Can I throw in a spanner?

 

I can show you posterisation from higher than 8 bit depth files (especially in the reds from my ....s). One thing that seems to be overlooked in the endless bit depth debate is that if the original data does not contain as much information as would be preferred then it may still exhibit unwanted effects such as posterisation. If sensors were as perfect as manufacturers would have us believe then 12 bit or higher and full information files would be great and an obvious way to go. However they are, unfortunately, not. So if a manufacturer knows this and compresses accordingly, the resultant 'loss' of data may be extremely small. Simply having greater bit depth is no guarantee of better colour 'quality'.

Share this post


Link to post
Share on other sites
Smaller battery or weaker processing?

 

How much battery power is saved by not running a compression algorithm on each image vs writing twice the amount of data to disk, I wonder? (This isn't a rhetorical question, incidentally - I seriously have no idea).

Share this post


Link to post
Share on other sites
Smaller battery or weaker processing?

More likely slower processing - and heat dissipation problems.

Share this post


Link to post
Share on other sites
Simply having greater bit depth is no guarantee of better colour 'quality'.

 

While that is true, it is no argument for lower bit depth, so we continue as we were...

Share this post


Link to post
Share on other sites
How much battery power is saved by not running a compression algorithm on each image vs writing twice the amount of data to disk, I wonder? (This isn't a rhetorical question, incidentally - I seriously have no idea).

 

The conversion would done using a fairly simple lookup table wouldn't it? If so there wouldn't be much processing involved. Like yourself this isn't my area of expertise, so I'd welcome to be proved wrong.

 

I think the decision could have been made on a combination of battery life and time taken to write the 16 bit images - especially if someone was using DNG+Jpeg.

 

Regarding reds, I've had a problem with very bright reds on every digital system I've ever used.

Share this post


Link to post
Share on other sites

I recall an early discussion paper from Leica in which they defended their decision on compression as being based on write speed and file size. These taking longer and being larger than M8 owners would tolerate, in their estimation. Sorry can't reference the exact source easily, but it was either on their website, or maybe in an LFI article.

 

Doug

Share this post


Link to post
Share on other sites
I recall an early discussion paper from Leica in which they defended their decision on compression as being based on write speed and file size. These taking longer and being larger than M8 owners would tolerate, in their estimation. Sorry can't reference the exact source easily, but it was either on their website, or maybe in an LFI article.

 

Doug

 

Write times are moot for me, as I never use Continuous mode. As for filesize, that is surely no longer an issue, now that much larger cards are an option.

 

I have a feeling that the change could be implemented quite easily. After all, the first M8s did not have the compression algorithm in place (before release, that is) - as tested by LFI in the weeks running up to the official release date.

 

One possible reason would be there will need to be some visible step up in quality when the M9 is finally released. Could it be that (surprise) that future camera would have no such in-camera compression? Just a possible thought. Otherwise I see no real reason for it...

Share this post


Link to post
Share on other sites
As for filesize, that is surely no longer an issue, now that much larger cards are an option.

 

Indeed its not, but of course when the M8 was originally released without SDHC support, the largest usable cards were 2 gig which I'm guessing would have meant a maximum of 100 photographs per card. There were 4 gig cards available - I used one - but they were the exception and didn't use the agreed SD specification.

Share this post


Link to post
Share on other sites

The actual mapping process - from 14 bit to 8 bit is trivial, accomplished by a lookup table. Although the mapping has the effect of multiplying by 4 and taking the square root, there are only 16384 possible inputs mapping to 256 possible outputs so the lookup table contains all possible results pre-calculated plus Leica can fiddle with the table as much as they want to, especially at the top and bottom ends.

 

I feel certain the decision to go with the 8 bit mapping was to do with battery power and card capacity, both of which are improving over time. I'm quite sure Leica could source and supply a higher capacity battery if they wanted to and I've shown how battery life could be extended by reducing the time-to-sleep.

 

I think what's handicapped them is the notion that the M needs a high burst depth and the magic number of 10 was plucked from thin air. I don't need that, I don't need continuous and if Leica said you can have uncompressed as an option but the burst depth is 5, not 10 and the battery life is halved (say), I think many of us would welcome it.

 

As the OP says, all that effort with the glass and they trip at the final hurdle...

Share this post


Link to post
Share on other sites

I concluded my investigation, by taking pictures with the lenscap on, at 1/4000 sec.

I took a serie of pictures at ISO 160, 320, 640, 1250 and 2500.

In this way I could calculate the ISO dependant Sensor Noise and the A/D converter noise.

The outcome of this is listed below.

 

 

Values for the sensor noise are all normalised at ISO 160, and numbers are in 14 bits values.

Well visible is that the M8 A/D converter compares very well against the other 3 cameras, is has the lowest noise of them all.

On the other hand, the sensor noise is the worst.

Total noise is calculated as: Sqrt((Sensor*ISO/160)^2 + (A/D) ^2)

 

This explains why at low ISO values the M8 is very good, noise wise, because the noise is almost only A/D noise.

At high ISO values, this becomes the opposite, this is where the sensor noise dominates.

That is why the M8 is not a very good high ISO performer.

Below is a plot of the Dynamic Range at 160 ISO and at 2500 ISO.

at ISO 2500, the darker the picture gets ( more to the left of the horizontal axis), the more dynamic Range ist lost compared to ISO 160.

The same plot for a 1Ds3 shows two almost parallel lines, which gives a huge advantage in the dark areas.

 

Hans

 

Share this post


Link to post
Share on other sites

As a last step, I just show below the difference it makes at ISO 2500 for the M8 and the 1D3.

The 1D3 has a little bit more A/D conversion noise, but the sensor noise is almost 1/3 of the M8 sensor noise.

At lower values of the luminance, the 1D3 has a significant better S/N (signal to noise) value, resulting in a Dynamic Range that is almost 1,5 EV larger.

 

At ISO 160, the two cameras are very much alike in S/N.

 

Hans

 

Share this post


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.

Loading...
Sign in to follow this  

  • Recently Browsing   0 members

    No registered users viewing this page.

×
×
  • Create New...