Sony versu Kodak, round one.

[ollypenrice]

I've always taken the view that my Kodak chipped cameras have been easy to work with. (Including Tom's and Yves', which are resident here, this involves 2xAtik 4000, 2xAtik 11000 and 1 x SXVH36. Thousands of hours.) I've also followed the orthodox view that if Sony made big chips I'd prefer them. Now I'm not entirely sure about that. I've done two images with a new (to me) Atik 460, chosen to give higher resolution than the 11 meg. It certainly does give this, but there's a but...

Let's follow a simple log stretch of 16x15 minutes of luminance from the 460/TEC140 F7. The linear stack looks like this. Once edge cropped there is a gap to the left of the histo.

Looks OK, though the stars are already bright. Next a log stretch. We reset the black point and give it a first log stretch...

This creates another huge gap to the left of the histo...

...which, when the black point is reset, gives us a background value of 6.

We stretch again...

 

Reset the black point for a background value of 23...

 

... and it is actually quite noisy.

Now what is unfamiliar to me is the behaviour of the histogram when stretched. Each stretch generates a huge gap to the left of the data line which, when corrected by resetting the black point, brings up the potential for background noise. When I stretch the Kodak data I get less of a gap and more of a stretch within the histo peak.

I can only suppose that the difference arises from the greater well depth of the 11 meg Kodaks with their 9 Micron pixels. I don't have the same target in the Kodaks but note the behaviour of the histo during similar log stretches of only 5x10 minutes Atik 11000 luminance data here:

Linear stack edge cropped.

 

First stretch...

Reset the black point:

which gives us

so we reset the balck point, stretch again...

Which gives us

When the black point is reset we have an image of relatively low noise noise, even in the dusty parts. This was just 5x10, remember, not 16x15!

So far, then, I cannot see that the Sony is really less noisy than the Kodak. Given the odd way that the Sony data stretches, I think the Kodak is actually winning, but I may be missing something.

I'm all ears regarding your thoughts.

In a subsequent thread I'll look into the issue of star colour. I'm not doing well with it in the Sony, be it said.

Olly

 

 

 

 

 

[steppenwolf]

That's an interesting one - I will be interested to see what happens with similar stretching using my Kodak 8300 sensor although I never stretch in this manner in the real world (and suspect that you don't either?) as I perform my stretches iteratively with a maximum 10 point adjustment each time.

Edited to add: a PS - I have never been a subscriber to the 'noisy Kodak sensor' club and standard calibration techniques work very well indeed!

[ollypenrice]

20 minutes ago, steppenwolf said:

That's an interesting one - I will be interested to see what happens with similar stretching using my Kodak 8300 sensor although I never stretch in this manner in the real world (and suspect that you don't either?) as I perform my stretches iteratively with a maximum 10 point adjustment each time.

Edited to add: a PS - I have never been a subscriber to the 'noisy Kodak sensor' club and standard calibration techniques work very well indeed!

Hi Steve,

I do stretch like this sometimes. I mostly use a log stretch like this (grey point slider to the left) for the RGB and, more likely, a custom curve for the luminance and narrowband. My NB stretches raise eyebrows but I got this from Adam Block, and Tom uses this kind of stretch routinely:

which gives

This is good for high contrasts where the signal is low.

However, my thinking on the Sony sensor is to go for a custom Curve whch combats the shallow well depth.  This would be a heavy lift at the bottom with an early flattening to avoid running out of range at the top. Something like this.

Purely experimental at this stage.

Olly

 

 

 

 

[swag72]

This isn't dissimilar to my thread a while back where I compared the a Sony and the Kodak chip...... In the real world I found the Sony to be MUCH noisier than the Kodak. 

I did put my findings on my website as well http://www.swagastro.com/real-world-comparison---kodak--sony-chip.html

[Davey-T]

Maybe there's some confusion between background noise and the snow storm of burnt out pixels on the 8300 ?

Dave

[andrew s]

I don't think it is noise as such. I think Olly is right that it is due to the well depth of the two chips. To compensate for this the gain of the Sony chip in the Atik 640EX is 0.27e-/ADU and that of the Atik 11000 is 0.92 e-/ADU a factor of about 3. Thus the output from the Sony chip is stretched by a factor of 3 by the camera in going from electrons to ADU so in effect it's dynamic range is 1/3 that of the Kodak. Only every 3rd ADU in the Sony chip represents a signal level whereas in the Kodak chip it is practically 1 to 1.

Your stretching  is showing this up as a comb effect.

This is down to the physical size of the pixels 4.54 x 4.54 for the Sony and 9 x 9 for the Kodak as this is the major factor that defined well depth. This is one area where binning can't help as the gain will not change. 

The read noise of the Sony is 5 e- compared to the Kodak's 13e-. 

All data is from the Atik site specs.

Regards Andrew

[gnomus]

Hi Olly.  I have noticed this business of stretching - finding that there is a gap - bringing in the black point - stretching again - finding another gap - and so on.  Two things occur to me.  Firstly, there have been times when I saw a left-sided gap in PS, but in PixInsight's HT there is either no gap or a much smaller gap before pixels start being clipped.  Could this be because the display in PS is 8 bit whereas HT is (I think at least) 16 bit?  Is it possible that there are pixels (albeit a small number) that are being clipped when you bring in the black point in PS?

The second point is me 'thinking out of me brains', and perhaps I am missing something.  But, what would happen if we had a perfect camera that had zero noise?  You point this at the night sky.  Bits of that sky will be 'empty'.  You decide, for artistic and other reasons, that you wish to have that bit of empty sky display at level 23 on your histogram.  If there is no noise, what would you expect to see to the left of that 23 point on the histogram?

[andrew s]

Producing gaps (combing) in the intensity histogram is very common in image processing and is one easy way to tell if an image has been manipulated.

Regards Andrew

[ollypenrice]

3 hours ago, gnomus said:

Hi Olly.  I have noticed this business of stretching - finding that there is a gap - bringing in the black point - stretching again - finding another gap - and so on.  Two things occur to me.  Firstly, there have been times when I saw a left-sided gap in PS, but in PixInsight's HT there is either no gap or a much smaller gap before pixels start being clipped.  Could this be because the display in PS is 8 bit whereas HT is (I think at least) 16 bit?  Is it possible that there are pixels (albeit a small number) that are being clipped when you bring in the black point in PS?

The second point is me 'thinking out of me brains', and perhaps I am missing something.  But, what would happen if we had a perfect camera that had zero noise?  You point this at the night sky.  Bits of that sky will be 'empty'.  You decide, for artistic and other reasons, that you wish to have that bit of empty sky display at level 23 on your histogram.  If there is no noise, what would you expect to see to the left of that 23 point on the histogram?

Good question. However, the background sky isn't black. There's Olber's Paradox, in which every line of sight should end in a star, but doesn't. This is resolved by the expansion of the universe but, never the less, many more lines of sight to a camera do end in stars than to the eye, and there is the scattering effect of the atmosphere. So I think a perfect camera could still record a genuine background of 23 (or whatever.) I don't believe that our deep images make background skies out of noise. Not primarily, at any rate. Starlight is light.

It's interesting that the Kodaks are much maligned, and even mocked in the case of my 'obsolete' 11 meg, when real world imaging does not support the Sony superiority theory - so far - for me. Maybe I should buy another Kodak 11 meg while I still can...

The Kodak snowstorm of hot pixels (which certainly exists but vanishes under stacking and calibration) does not seem to be a good indicator of final noise in a processed image.

Olly

[gnomus]

25 minutes ago, ollypenrice said:

Good question. However, the background sky isn't black. There's Olber's Paradox, in which every line of sight should end in a star, but doesn't. This is resolved by the expansion of the universe but, never the less, many more lines of sight to a camera do end in stars than to the eye, and there is the scattering effect of the atmosphere. So I think a perfect camera could still record a genuine background of 23 (or whatever.) I don't believe that our deep images make background skies out of noise. Not primarily, at any rate. Starlight is light.

It's interesting that the Kodaks are much maligned, and even mocked in the case of my 'obsolete' 11 meg, when real world imaging does not support the Sony superiority theory - so far - for me. Maybe I should buy another Kodak 11 meg while I still can...

The Kodak snowstorm of hot pixels (which certainly exists but vanishes under stacking and calibration) does not seem to be a good indicator of final noise in a processed image.

Olly

Thanks for responding.  I quite agree that the background sky isn't 'black' and that it would be a mistake to produce a picture where the background was set at 0.  But (ignoring for the moment dark nebulae) what is blacker?  And if nothing is blacker, then what would we expect to see in a histogram from a perfect no-noise camera where you have set the background to be 23, 20, 15, or whatever?  Is it possible that what we see to the left of our chosen black point for the empty sky is largely camera noise, in which case a gap might not necessarily be a bad thing?  In the last couple of pictures I have done, I got to a point where I had the sky set where I wanted it (around the 22 mark) and I had a 'full' histogram that went all the way to the left edge.  I then ran Noel's Deep Sky Noise Reduction.  The picture looked smoother although the tones had not changed significantly.  And I noticed a gap had appeared at the left edge.  Like you I feel that I should have a histogram that goes all the way to the left edge, but perhaps this is not the case.  I was happy with my picture before I ran the NR.    

Also, I am making no great plea for Sony cameras.  I have one of them, but I also have two KAF 8300 cameras (and have just bought a third).  

[gnomus]

Just to follow up that last post, I very quickly took a couple of screen shots of an earlier version of my Leo Triplet image.

The first image is what I get when I bring up the levels dialogue.  I made no levels adjustment.  You can see that the histogram does go all the way to the left edge:

I then ran Deep Space NR and brought up levels again.  Now you can see a clear gap (it took 5 points in the black point window to eliminate this gap):

 

Incidentally, this is the KAF 8300 chip.

 

[ollypenrice]

1 hour ago, gnomus said:

Just to follow up that last post, I very quickly took a couple of screen shots of an earlier version of my Leo Triplet image.

The first image is what I get when I bring up the levels dialogue.  I made no levels adjustment.  You can see that the histogram does go all the way to the left edge:

I then ran Deep Space NR and brought up levels again.  Now you can see a clear gap (it took 5 points in the black point window to eliminate this gap):

 

Incidentally, this is the KAF 8300 chip.

 

OK, firstly let's think about the background sky level from our imperfect cameras. However imperfect they may be, they consistently and successfully distinguish between background sky and obscuring dust, which they render much darker.  This is clear proof that our background skies are not made of noise.

Regarding those damned gaps to the left of the data line (which for the moment I will call Sony Gaps!) well, they are most upsetting. It is indeed very tempting to leave them there since the image looks best when you do. But it means that you are not using the full dynamic range of the screen version. This is anathema to me, I have to say. What I want to do is somehow persuade the data to fill the full screen brightness range. I don't doubt that, with the 460, I could shoot enough data to kill the gap but then the cores would be blown. Multiple sub lengths? It may come to that. Or some kind of custom curve as I mentioned in my first reply to Steve Richards.

What is Noel doing with his NR? I suspect that he's identifying errant dark pixels and cranking them up to the value of their neighbours. This will leave a gap, no? In images without dark nebulae I often do my NR myself, rather than using a 'one size fits all' routine. I zoom in to pixel scale and 'colour select' (it works in greyscale) the abnormally dark pixels which I then hit with the Median filter, which gives them the average value of their neighbours. I also pin the higher background value and give the curve a gentle lift below that to squeeze the histo in the background. Ditto, this will give a gap because we have eliminated the darkest pixels. Do not do this when imaging the Barnard catalogue!!!

lly

 

 

[RichLD]

I am looking to buy another KAF 8300 camera. Based on my own experience I think it's an amazing chip which calibrates brilliantly. I really can't understand all this fuss about it being insensitive and noisy - it, along with the other Kodak chips takes fantastic AP images, nuff said 

[ollypenrice]

5 hours ago, RichLD said:

I am looking to buy another KAF 8300 camera. Based on my own experience I think it's an amazing chip which calibrates brilliantly. I really can't understand all this fuss about it being insensitive and noisy - it, along with the other Kodak chips takes fantastic AP images, nuff said 

At this early stage I'm heartily inclined to agree!

Olly

[andrew s]

Just remember that it is not about Sony v Kodak as such it is about the specs of the chips and the quality of the implementation.

As an example the well depth the Kodak KA 11002 is 60,000 the KAF-8300 25,500 and the ICX694 18,000. In effect this means that the KA 11002 has a resolution of just under 16 bis while the ICX694 has in effect only slightly more than 14 bits and the KAF 8300 just under 15 bits. Just because they have 16 bit A to D converters does not change this.

Other things to consider is QE, read noise etc. etc. It pays to look at all the factors and not just Sony v Kodak.

I know looking at resultant images can help but only if the processing is done with the same skill and understanding in all cases.

For my spectroscopy work the ICX694 performs better than the KAF 8300 (I have both) but that is because of the specific use (QE & low read noise are a big factors) however, it may not be the case generally.

Regards Andrew

[ollypenrice]

I have added about 1.6 of the 2.0 missing mm to the chip distance from the TEC 140's flattener and will be interested to see if the Atik 460 plays better in this configuration. I can't say I'm optimistic but who knows? There is no easy way to obtain the perfect chip distance but on such a small chip I can't think it matters.

We have a nice test target ce soir so let's see what it gives!

Olly