Can someone help out by commenting or posting what a good flat looks like (I use a Canon 400d) - in particular how bright (or dim) the corner versus the edges should be.
I image at ISO 800 and took some flats with a white T shirt over the main OTA lens cap just around sunset with the scope pointing high in a patch of blue sky (taking images of duration 1, 2, 4, 6, 8, 10 , 12 and 16 seconds).
How do I tell what an optimal series of flats should look like please! I presumed you want it semi bright in the centre and fading to dull (not dark) at the edged?
I always have recommended a flat that has the histogram approx one third the way along.
So whatever exposure will give you this type of histogram is fine.
The pefect flat should be just that-perfectly flat.
The reality is that there is usually some light drop off towards the corners of the frame resulting in the dimmer corners seen on the flat you describe. The average histogram for the entire frame should be about half the total possible for your sensor. It should not be much more than that as the antiblooming feature (if present) will make the sensor response non linear above this level and the resulting flat won't be as accurate.
Andrew - when you say 1/3 of the way along - you mean from the left (dark) end don't you? So you expect you shot to look dim, not dark nor too bright (i.e. like a picture of the sky towards late sunset).
Terry - when you say a perfect flat should be perfectly flat - what do you mean? Saying a perfect egg should be a perfect egg - doesn't help a newbie trying to understand that definition.
Also when you say the average histogram should be half the total possible - I bet you're right - but what does that look like - i.e.
If its done right does the Histogram mountain in Photoshop CS2L
1. 1/2 the height of the box, or
2. half the width of the box or
3. half way along the bow or
4. something else entirely!
I guess I am asking for a lay persons guide to knowing if I got it right!
Andrew - when you say 1/3 of the way along - you mean from the left (dark) end don't you? So you expect you shot to look dim, not dark nor too bright (i.e. like a picture of the sky towards late sunset).
Exactly, you read the histogram from left to right so 1/3rd from the left (dark).
That's how I've done it.
Quote:
Originally Posted by g__day
Also when you say the average histogram should be half the total possible - I bet you're right - but what does that look like - i.e.
If its doene right does the Histogram mountain in Photoshop CS2L
1. 1/2 the height of the box, or
...
...
I guess I am asking for a lay persons guide to knowing if I got it right!
Just to clarify a point on histograms.
The height has nothing/little to do with what we're concerned about.
It only indicated how many of the pixels are at that grey level (out of a possible 256).
Each channel (RGB) is represented by 256 grey levels (0-255) in 8 bit mode and so on.
The height can go off scale vertically but it's the horizontal line we need to keep an eye on and tells us when it's clipping.
Andrew - when you say 1/3 of the way along - you mean from the left (dark) end don't you? So you expect you shot to look dim, not dark nor too bright (i.e. like a picture of the sky towards late sunset).
Terry - when you say a perfect flat should be perfectly flat - what do you mean? Saying a perfect egg should be a perfect egg - doesn't help a newbie trying to understand that definition.
Also when you say the average histogram should be half the total possible - I bet you're right - but what does that look like - i.e.
If its doene right does the Histogram mountain in Photoshop CS2L
1. 1/2 the height of the box, or
2. half the width of the box or
3. half way along the bow or
4. something else entirely!
I guess I am asking for a lay persons guide to knowing if I got it right!
Better still - can someone please post a sample dark and the histogram demonstrating how it should look. I've just built a lightbox and would appreciate any tips on how to use it
Here's a flat I dug up from a while ago, probably won't show up too well since it's been converted for the web but the original shows a few dust motes (top left, bottom right).
This was using the TOA-130 and the 20Da.
Andrew - when you say 1/3 of the way along - you mean from the left (dark) end don't you? So you expect you shot to look dim, not dark nor too bright (i.e. like a picture of the sky towards late sunset).
Terry - when you say a perfect flat should be perfectly flat - what do you mean? Saying a perfect egg should be a perfect egg - doesn't help a newbie trying to understand that definition.
Also when you say the average histogram should be half the total possible - I bet you're right - but what does that look like - i.e.
If its done right does the Histogram mountain in Photoshop CS2L
1. 1/2 the height of the box, or
2. half the width of the box or
3. half way along the bow or
4. something else entirely!
I guess I am asking for a lay persons guide to knowing if I got it right!
No 3 is correct. Halfway along the box.
In an ideal world (with perfect optic) an image taken of a flat even scene would result in every pixel having the same levels. This is what you want to achieve. The role of a flat field is to simulate this.
When you take an image through an optical train there will be some variation in the illumination across the field caused mostly by dust and vignetting in the lens (telescope).
If you divide the image by a "flat" field this will hopefully correct for the variation in illuimination in your image. The problem with this is the introduction of noise into the image from the flat field. This need to be limited as much as possible. If the numbers that you are dividing by are big then there is less noise added to the image compared to dividing by small numbers. The size of the numbers depends on the histogram in the flat- hence this explanation. If your CCD has a maximum count of ~32000 then making most of the numbers about half this seems to give the best result. This is achieved by having the histogram about half way along the box. What your flat actually looks like doesn't matter as long as it works.
A test is to take a number of flat images and keep one separate. Make a master flat using whatever tool your software provides. Then divide the flat frame you kept separate bu your master flat. The result should be a perfectly image with as little noise as possible.
Hope this is helpfull.
Thanks for that - will try to nail a shot with the histogram square in the middle.
A related query springs to mind. If I have the picture and I have the flat, is it easy to do the correction processing in Photoshop CS2 (rather than DeepSkyStacker - which always leaves the colours a bit wierd) and what are the steps involved if anyone knows?
Many thanks, this is great,
Matthew
PS
When you say if your CCD has a MAX count of 32000 - is that the well depth of each pixel? If I am on a Canon 400D with 12 bit A/D converters to I say the pixel depth is 2 ^ 12 = 4096 so aim for 2048 no more and this is revealed in Photoshop CS2 somewhere?
When you say if your CCD has a MAX count of 32000 - is that the well depth of each pixel? If I am on a Canon 400D with 12 bit A/D converters to I say the pixel depth is 2 ^ 12 = 4096 so aim for 2048 no more and this is revealed in Photoshop CS2 somewhere?
Yes that is correct. The ~32000 was just an example because it is the well depth of my camera at 14bit.
When I look at my histrogram in PS CS2 - I see three curves for each colour channel - will post an example tonight - the three don't overlap much - is that an indication that colour balance is way out on my flat and I should re-balance colour at the camera settings and do the flat again?
Well the filters job is to slow down / stop some light and let other light in, so your histogram will show that. I get the same sort of thing with my filter. Light colours are at different frequencies so it stands to reason that some colours are slowed down. Make a Custom White Balance for your camera with the filter in and then see what you get.
Else when you have the processed image in PS, do a levels on each colour to align them. You might need to curve the whole show first to seperate the colours to a point where you can see the difference, depending on the image data.
That's basically what I do - a levels first overall, slight curves, then individual play with RGB levels so one colour isn't flooding the rest, then alot more subtle curves and reseting the black point, finally I go into colour correction to try and get my subject looking how I think it should.