North American and Pelican Bicolor

[tooth_dr]

15 minutes ago, gorann said:

I also noticed the posterization in the first version but it looked better in the last. It is something to be aware of when stretching images. At least in PS the curves function works only in 256 levels (2 to the power of 8 ) so it is important to make many small stretches not to loose dynamic range as the original 16 bit image contains much more than 256 levels (actually 2 to the power of 16 so 65536 levels). I usually do about 10 iterations of stretching. Not sure how PI works here. By the way Rodd, this is posterization (the name coming from its use in posters, particularly by An Warhol):

This is interesting.  I didnt know the reason for doing multiple stretches, and quite often simply didnt bother wasting time, and did one big stretch.  Now that I understand the negative impact I will do many smaller interations.

[Rodd]

7 minutes ago, tooth_dr said:

Next time you pour some milk in your tea have a read at the label ?

 

?

[Rodd]

23 minutes ago, gorann said:

so it is important to make many small stretches not to loose dynamic range

My first stretch is usually BIG.  Maybe time to rethink

Rodd

[gorann]

Small stretches guys, always remember, only small stretches. We do not want stretch marks in our images....

[wimvb]

1 hour ago, Rodd said:

Ah--so more data is better even if you can't really see it

Rodd

Posterization is  hidden in every image. But if you use average stacking, you can reduce it. Eg if a intensity value in reality is 0.33, it will, due to noise, sometimes register as 0, and sometimes as 1 in your subs. By averaging subs, you will get the real value 0.33 in the stacked image, even though each sub registered 0 or 1 (or even 2 occasionally as an outlier). Otoh, median stacking will always give 0 or 1 (odd number of subs) in the stacked image. That's why you shpuld always choose average as the stacking method.

[tooth_dr]

28 minutes ago, wimvb said:

Posterization is  hidden in every image. But if you use average stacking, you can reduce it. Eg if a intensity value in reality is 0.33, it will, due to noise, sometimes register as 0, and sometimes as 1 in your subs. By averaging subs, you will get the real value 0.33 in the stacked image, even though each sub registered 0 or 1 (or even 2 occasionally as an outlier). Otoh, median stacking will always give 0 or 1 (odd number of subs) in the stacked image. That's why you shpuld always choose average as the stacking method.

I use sigma clipping, is this wrong?

[wimvb]

1 hour ago, tooth_dr said:

I use sigma clipping, is this wrong?

No, sigma clipping just takes away the outliers. Suppose that you have a bright pixel in a dark nebula. The pixel value is 255 (bright), while the nebula is 10. If you stack 10 subs, all with 10 in pixel value, except one with 255, then the average will be (9*10 + 1*255)/10 = 345/10 = 34.5 But if you use sigma clipping, all values that are too far from the average will be removed before the final average is calculated: 9*10 / 9 = 10, since the 255 value is removed.

[wimvb]

3 hours ago, Rodd said:

My first stretch is usually BIG.  Maybe time to rethink

Rodd

 

3 hours ago, tooth_dr said:

This is interesting.  I didnt know the reason for doing multiple stretches, and quite often simply didnt bother wasting time, and did one big stretch.  Now that I understand the negative impact I will do many smaller interations.

3 hours ago, gorann said:

Small stretches guys, always remember, only small stretches. We do not want stretch marks in our images....

Guys, guys, guys. In PixInsight, you can stretch the living daylight out of an image. PS works in either 16 bit integer mode or 32 bit integer mode. This means that pixel values are always whole (integer) numbers: 0, 1, 2, ..., 255, 256, ... 65535 for 16 bit images. If a weak dark nebula only "occupies" the pixelvalues 0 ... 10, it will only have 10 levels of intensity variation.

PixInsight on the other hand, works with 32 bit floating point accuracy. This means that pixel values are represented by decimal numbers 0 ... 1. This opens up a more continuous scale of pixel values. While this may not be fully utilized in a stacked image, it can be an advantage during processing.

Here's an other example of posterization:

[tooth_dr]

1 hour ago, wimvb said:

 

Guys, guys, guys. In PixInsight, you can stretch the living daylight out of an image. PS works in either 16 bit integer mode or 32 bit integer mode. This means that pixel values are always whole (integer) numbers: 0, 1, 2, ..., 255, 256, ... 65535 for 16 bit images. If a weak dark nebula only "occupies" the pixelvalues 0 ... 10, it will only have 10 levels of intensity variation.

PixInsight on the other hand, works with 32 bit floating point accuracy. This means that pixel values are represented by decimal numbers 0 ... 1. This opens up a more continuous scale of pixel values. While this may not be fully utilized in a stacked image, it can be an advantage during processing.

Here's an other example of posterization:

When I open the curves dialogue box and stretch my CCD images in PS they look like the image on the left above. But once I click ok and confirm the changes it looks like the one on the right. Any ideas why this is?

[Rodd]

1 hour ago, wimvb said:

Guys, guys, guys. In PixInsight, you can stretch the living daylight out of an image. 

So it does not matter--big or small?

Rodd

[wimvb]

13 minutes ago, Rodd said:

So it does not matter--big or small?

Rodd

I would like to say: no. If the "PRODUCT" of both stretches is the same as the single aggressive stretch, then it doesn't make a difference:

Out = stretch_1 x in

Out = stretch_2 x in

Out = stretch_2 x stretch_1 x in = BigStretch x in

In general, this will not be the case. But the difference isn't related to the format of the input image, only to the mathematical formula of the stretches.

[wimvb]

20 minutes ago, tooth_dr said:

When I open the curves dialogue box and stretch my CCD images in PS they look like the image on the left above. But once I click ok and confirm the changes it looks like the one on the right. Any ideas why this is?

I know too little about PS, and how it displays images. The act of stretching, shouldn't smooth out the image. There may be an actual step in the intensity values, that is not because of posterization. Stretching must preserve such a step, or blurring will occur. Posterization effects in 16 or 32 bit images are usually very small. What you see, may be how PS calculates what to show on screen.

[Rodd]

7 minutes ago, wimvb said:

I would like to say: no. If the "PRODUCT" of both stretches is the same as the single aggressive stretch, then it doesn't make a difference:

Out = stretch_1 x in

Out = stretch_2 x in

Out = stretch_2 x stretch_1 x in = BigStretch x in

In general, this will not be the case. But the difference isn't related to the format of the input image, only to the mathematical formula of the stretches.

Phew...For a minute there I thought I was going to have to redo them all!

Rodd

[wimvb]

No need, let the image decide. If it looks good, it is good, no matter what "theory" says.

My more or less standard procedure is one fairly hard histogram stretch with the midpoint towards the left, and the blackpoint towards the foot of the histogram. The peak if the histogram ends up at about 1/4 - 1/3 in from the left. After that I stretch with curve transformation for fine control. For rgb I may start with arcsinh or masked stretch. This boosts colour, at the cost of contrast.

[gorann]

For PS I read somewhere some years ago that it was important to do many small stretches to avoid posterization and I have also experienced ending up with a posterized image after stretching and that was fixed when I started over with more smaller stretches. But PI is so different from PS that it may not be an issue there. I assume we should trust Wim on this but if in doubt it could be easily found out by comparing the results of one or a few big stretches with that of many small. An interesting question is why Rodd ended up with posterized dark nebulosity after PI processing. From what I can see on the web, the risk of posterization is great when stretching 8 bit images and smaller in 16 bit images, which of course makes sense. Many CMOS cameras produces 12 or 14 bit images so the problem could be worse in CMOS than CCD.

Here is some info I now found on the web:

http://facweb.cs.depaul.edu/sgrais/image_posterization.htm

[Filroden]

8 minutes ago, gorann said:

From what I can see on the web, the risk of posterization is great when stretching 8 bit images and smaller in 16 bit images, which of course makes sense. Many CMOS cameras produces 12 or 14 bit images so the problem could be worse in CMOS than CCD.

This is one of the reasons for stacking many shorter exposures with CMOS, so you can recover dynamic range beyond its native 10 or 12 bits.

[Rodd]

2 hours ago, gorann said:

An interesting question is why Rodd ended up with posterized dark nebulosity

I think the posturization you are referring to is actually an injudicious use of curves--I wanted to lighten the "black" regions and used curves to brighten them.....obviously a mistake...one that I did not repeat with such force in the last version.  

Rodd

[Rodd]

2 hours ago, gorann said:

Here is some info I now found on the web:

Interesting what is said about processing in 16 bit mode with 8 bit data near the end.  

Rodd

[gorann]

1 hour ago, Rodd said:

Interesting what is said about processing in 16 bit mode with 8 bit data near the end.  

Rodd

Yes. The risk of posterization is of course greatest in stretching the very darkest end of the histogram, for example when trying to get to that faintest of dust. At some point here there will enevitable end up as posterization

[ollypenrice]

The 'one big stretch' has its place. If using a simple log stretch I'll do multiple small iterations but I often give Ha a curve I saw in, I think, an Adam Block video years ago. I use it on Ha when looking to extract the faintest signal.

In my imaging tutorials I often demo this curve saved as a layer and then do another layer using the conventional small-step grey point stretches so that the results can be compared by blinking from one to the other. So far as I can see the 'one big stretch' cannot be replicated in small iterations.

The difference is that the one big stretch gives much higher contrasts, partcularly in the faint signal. It generally gives a more dramatic look while the usual method looks softer, fluffier and more natural. Most of our guests prefer the hard stretch but I don't think Steve Richards was convinced when I showed it to him! ?

Because I'm usually wanting to add Ha to red, in which anything faint will struggle to register over the basic red and much of the contrast will be diluted in the blending anyway, I like the One Big Stretch. The posterization seen in the screen grab appears only in the preview. Once applied I see no posterization at all. It does give bigger stars, of course, but they tend to be smaller than the stars in red so this doesn't matter. Someone imaging in pure NB would be concerned about this, though.

I use the One Big Ha Stretch in at least 90% of my images.

Olly

Edit: I should add that the One Big Stretch is usually followed up by some conventional small ones, sometimes as selectively blended layers to avoid saturation..

[Rodd]

14 minutes ago, ollypenrice said:

n my imaging tutorials I often demo this curve

Olly--One of the hardest things for me to do is look at the PS curve (like the one you showed) and equate it to PI curves--its almost like they are opposites somehow--maybe there is a scale difference involved.  I have yet to attempt to use PS (though I do now possess the dark tome).  My fear is....well, I have many, but staring at the screen and not having the slightest idea what to do is fairly high on the list.  I know you are at least familiar with PI--not sure if you have used curves in PI--but opening curves is just a matter of clicking on it.  Its even a box like it is in PS.  Are PS and PI curves dissimilar in some way--or am I not looking at it correctly?

Rodd

[ollypenrice]

18 minutes ago, Rodd said:

Olly--One of the hardest things for me to do is look at the PS curve (like the one you showed) and equate it to PI curves--its almost like they are opposites somehow--maybe there is a scale difference involved.  I have yet to attempt to use PS (though I do now possess the dark tome).  My fear is....well, I have many, but staring at the screen and not having the slightest idea what to do is fairly high on the list.  I know you are at least familiar with PI--not sure if you have used curves in PI--but opening curves is just a matter of clicking on it.  Its even a box like it is in PS.  Are PS and PI curves dissimilar in some way--or am I not looking at it correctly?

Rodd

I must try PI Curves because it didn't have them when I started. Ps and PI stretches are actually very similar, in practice. I never find much diffeence between a PI log stretch and a Ps log stretch. PI does give more information and more precise control but I...erm... use Ps!

The guy who could fix a puncture on a motorbike faster than anyone else I ever met used a screwdriver instead of tyre levers...

Olly