I need help with flats

[chrisvdberge]

I really need to start making flats, but am a bit confused on this one. I use a dslr and right now I'm trying to see if I can make flats using a t-shirt over an iPad (on white screen).
If I use the aparture setting (as recommended as far as I know) so that the camera decides the shutterspeed, I get a very dark image.
If I use manual setting and set the shutterspeed to such a length so that I don't clip any pixels, I get a completely grey image like this one;


If I stretch it the vignetting and dust becomes visible, however I don't think I should stretch a flat right? But somehow I do see people posting 'white flats'. So how do I achieve this??

Same image as above with an Auto ScreenTransferFunction


Do I need to stretch it after all?

[pfile]

you should only debayer and stretch the flat to see what it looks like. when applied during calibration the flat should be raw and linear like all the other frames.

STF is usually enough to see what's going on with the flat.

i usually do +2EV on DSLR flats because the AV mode tends to underexpose them a lot. normally you stretch every terrestrial image so what the camera computes is fine for normal photography, but not for flats. generally you try to get flats into just about 1/2 well depth for your camera; you want to have a decent amount of signal but you need to stay within the linear response zone of the CCD. 1/2 well depth is a good compromise for most sensors.

rob

[chrisvdberge]

thx pfile, does that mean I need to keep the histogram left of the middle? (not sure what you mean with 1/2 well depth). And so it is ok for the flat to be completely black (as linear image)?

furthermore; does the flat need to be shot in greyscale or can I just shoot it as color and not debayer it?

[pfile]

most newer DSLRs have 14-bit A/D converters. PI brings these 14-bit values into a 16-bit space, so a completely overexposed flat loaded in PI will have it's histogram crammed up against the value 0.25. so 1/2 well depth would be 0.125 in PI's histogram. other programs may be different (they may scale the data) so it's best to find out where the histogram lies when you feed the program a completely overexposed image.

if your camera has a greyscale setting, never use it. that will result in the camera debayering the image and converting it to greyscale before it's handed to you. that makes the file useless as a flat. just shoot your flats as you normally would.

some confusion stems from the fact that *all* CCD images are mono images. when using a mono camera, you make 3 images (R, G, B). with an OSC (one-shot color) like a DSLR, the filters are basically glued right on top of the sensor. there 3 filters per square block of 4 pixels, and there are 2 green pixels per block. if you open one of these images as "raw cfa" you are looking at the data exactly as recorded by the sensor. debayering goes in and "deinterleaves" the pixels into 3 planes and assigns the color R to one plane, G to the other and B to the last.

rob

[chrisvdberge]

thx, i've shot some lights and flats yesterday and will see if I can calibrate them to good results
I was a bit anxious of the fact that a lot of pixels were rejected when creating the masterFlat, but it seems that I need to have this setting on right?
Comparing 2 frames after initial calibration shows that I got rid of the vignetting, but a bit strange noise pattern and a LOT less signal. But I will make to complete stacks to make a fair comparison first before drawing any conclusions


as for the histogram placement; if you underexpose the flat, will PI correct this somehow or will you just have a poor quality result?

[MikeOates]

Chris,

The problem with DSLR images is they are far noisier than CCD's, so they do have to be pre processed and post processed differently.

I found that exposing the flats so that the peak is at about 3/4 across from the left. This it make sure the flat is not under exposed which will add noise to your image after calibration and that the last thing the calibration operation should be doing. Doing this I got good results. But please note: The histogram I am refering to is the one shown in Lightroom, this is almost certainly not the same as in PI and may be different on back of the camera. Once you have established an exposure, just use that for all your images and to do that your best using the camera in fully manual mode.

Also more bias frames are needed, typically 100-200 frames, I used 150.

Regards,

Mike

[pfile]

like any stack of images if you use enough subexposures then the SNR will increase as the square root of the number of frames. so if your flats have low signal, well, the SNR will be improved by using a lot of them. but i figure why start with poor data in the first place?

i agree with mike that you should use a ridiculous number of bias frames with canon DSLRs. this can help prevent injecting more banding noise into the calibrated subs.

rob

[MikeOates]

i agree with mike that you should use a ridiculous number of bias frames with canon DSLRs. this can help prevent injecting more banding noise into the calibrated subs.

rob

Yes I should have mentioned that, since I started taking 150 bias frames with my Canon 500D (full spectrum mod) I no longer had to use the remove canon banding script, all banding was removed during calibration with 150 bias frames.

Mike