Jump to content

  •  

CNers have asked about a donation box for Cloudy Nights over the years, so here you go. Donation is not required by any means, so please enjoy your stay.

Photo

Calibrating Out Fixed Pattern Noise in CMOS Lights using PixInsight

  • Please log in to reply
70 replies to this topic

#1 AXAF

AXAF

    Apollo

  • -----
  • topic starter
  • Posts: 1,050
  • Joined: 15 Sep 2004
  • Loc: MA

Posted 04 January 2020 - 02:58 PM

I've been noticing that I am not able to completely calibrate the fixed pattern noise from my images. For reference, I am processing through

PixInsight. Frames are collected using a a ZWO 183MM Pro sensor. My flat fields are taken using a Spike-a Flat Fielder. I generate a master

flat per filter using sixty exposures at the same gain and offset values as the light frames, themselves. The histogram of a single flat field

looks reasonable to me. I also apply the necessary Bias master and Dark frame master when calibrating the raw flat field frames off of the camera.

 

Looking at an individual, calibrated light frame, it's very difficult to detect any residual FPN. But if I aggregate many calibrated

light frames, e.g. two hundred, the FPN becomes perceptible.

 

I basically am following the procedures outlined in Charles Bracken's book, "The Deep-Sky Imaging Primer, 2nd Ed." Since I am

working with CMOS, I never select the Optimize option at any point in either generating the flat field master, or the calibrated

target lights.

 

Here is an overly stretched example of the Leo Triplet with about 400 calibrated light frames to illustrate my point.

Attached Thumbnails

  • LeoTriplet.PNG

Edited by AXAF, 04 January 2020 - 03:01 PM.


#2 AXAF

AXAF

    Apollo

  • -----
  • topic starter
  • Posts: 1,050
  • Joined: 15 Sep 2004
  • Loc: MA

Posted 04 January 2020 - 02:59 PM

Here is a single luminance flat field frame. The histogram looks OK to me, is it not?

Attached Thumbnails

  • Flats_L.PNG


#3 Cfreerksen

Cfreerksen

    Surveyor 1

  • *****
  • Posts: 1,670
  • Joined: 18 Aug 2018
  • Loc: Tooele, Ut

Posted 04 January 2020 - 03:11 PM

Dither.

 

Chris


  • happylimpet and ks__observer like this

#4 AXAF

AXAF

    Apollo

  • -----
  • topic starter
  • Posts: 1,050
  • Joined: 15 Sep 2004
  • Loc: MA

Posted 04 January 2020 - 03:14 PM

Dither.

 

Chris

What does dithering have to do with the variation in pixel sensitivity, ie. FPN? Nothing. Thanks.



#5 kyle528

kyle528

    Viking 1

  • *****
  • Posts: 713
  • Joined: 02 Feb 2019
  • Loc: SE Ohio

Posted 04 January 2020 - 03:18 PM

What does dithering have to do with the variation in pixel sensitivity, ie. FPN? Nothing. Thanks.

You're wrong. Dithering has EVERYTHING to do with removing FPN. 


  • ks__observer likes this

#6 jhayes_tucson

jhayes_tucson

    Fly Me to the Moon

  • *****
  • Posts: 7,388
  • Joined: 26 Aug 2012
  • Loc: Bend, OR

Posted 04 January 2020 - 03:19 PM

First off, it's very hard to visually identify FPN unless there is an obvious pattern.  Remember that FPN is due to variations in sensor responsivity as a function of position.  On the flat that you posted, the most obvious FPN visible are the horizontal streaks.  For your sensor, FPN will mostly look like spatially random variations across the signal.  I'll also mention that since FPN is a multiplicative modulator (just like vignetting,) it calibrates out of your data by flat fielding (which will add a small amount of noise to the output.)  When I look at your calibrated frame, it does appear that the FPN is mostly being removed.

 

Second, the dust mote in your calibrated image clearly shows that flat calibration may not be working properly.  I say "may" only because your flat frame shows a lot of dust motes that are clearly being calibrated out.  Was that last remaining dust mote a new one that appeared between when you took the data and the flats?  One other thing:  The most significant thing that can screw up flat correction is stray light.  Stray light is additive and not multiplicative so it won't be removed and it may screw up your flat correction if you are getting strays in the flat data.  So be very careful about stray light!

 

Finally, Chris is correct that dithering will help to further minimize the effects of non-spatially random FPN.  If you dither AND calibrate each frame, the most obvious effects of non-spatially random FPN should be rendered insignificant.

 

John


Edited by jhayes_tucson, 04 January 2020 - 03:20 PM.

  • AXAF, psandelle, jeffweiss9 and 4 others like this

#7 AXAF

AXAF

    Apollo

  • -----
  • topic starter
  • Posts: 1,050
  • Joined: 15 Sep 2004
  • Loc: MA

Posted 04 January 2020 - 03:26 PM

First off, it's very hard to visually identify FPN unless there is an obvious pattern.  Remember that FPN is due to variations in sensor responsivity as a function of position.  On the flat that you posted, the most obvious FPN visible are the horizontal streaks.  For your sensor, FPN will mostly look like spatially random variations across the signal.  I'll also mention that since FPN is a multiplicative modulator (just like vignetting,) it calibrates out of your data by flat fielding (which will add a small amount of noise to the output.)  When I look at your calibrated frame, it does appear that the FPN is mostly being removed.

 

Second, the dust mote in your calibrated image clearly shows that flat calibration may not be working properly.  I say "may" only because your flat frame shows a lot of dust motes that are clearly being calibrated out.  Was that last remaining dust mote a new one that appeared between when you took the data and the flats?  One other thing:  The most significant thing that can screw up flat correction is stray light.  Stray light is additive and not multiplicative so it won't be removed and it may screw up your flat correction if you are getting strays in the flat data.  So be very careful about stray light!

 

Finally, Chris is correct that dithering will help to further minimize the effects of non-spatially random FPN.  If you dither AND calibrate each frame, the most obvious effects of non-spatially random FPN should be rendered insignificant.

 

John

Excellent explanation. Thanks, John. -Gary
 



#8 Tapio

Tapio

    Fly Me to the Moon

  • -----
  • Posts: 6,073
  • Joined: 24 Sep 2006
  • Loc: Tampere, Finland

Posted 04 January 2020 - 03:30 PM

Although I don't see too much FPN I say also, dither.
  • psandelle likes this

#9 freestar8n

freestar8n

    Vendor - MetaGuide

  • *****
  • Vendors
  • Posts: 11,457
  • Joined: 12 Oct 2007
  • Loc: Melbourne, Australia

Posted 04 January 2020 - 05:11 PM

The fact that you can see the one dust mote indicates the flats are not well matched with the lights.  People worry about the exposure of flats, but as long as they aren't exposed too much and as long as you have enough exposures they should be ok - but only if they match the illumination of the lights.  I would try a clear sky or t-shirt flat and see if it works better.  If it is bright outside when you take the flats, make sure the imaging train is well shielded from light.

 

It's nice to see people talking more about pattern noise and dithering - but it's important to distinguish fixed pattern noise, which is identical in each frame, from PRNU due to variations in pixel response.  The latter is a form of pattern noise but the resulting noise in the image will depend on the scene being imaged - which makes it fundamentally different from FPN - even though there is a pattern to it.

 

Both FPN and PRNU would be helped by dithering - but the presence of the dust mote after calibration suggests the flats and calibration procedure are not optimal and should be addressed first.  

 

Frank


  • psandelle, happylimpet, Michael Covington and 1 other like this

#10 Michael Covington

Michael Covington

    Author

  • *****
  • Posts: 7,639
  • Joined: 13 May 2014
  • Loc: Athens, Georgia, USA

Posted 04 January 2020 - 05:19 PM

Let's revisit the question of flat darks.  I gather you are using bias frames but not flat darks.  With many cameras, the two are essentially the same thing.  But I am wondering if this calibration step is not being applied to the flats.  If so, you end up with calibrated flats that have the right arrangement of dust motes but not enough amplitude (contrast) and have less effect than they ought to.  Can you tell me the full workflow and software?

Dithering would further improve your pictures, probably, but it wouldn't cure the problem we're looking at, so I think we want to get the process working well without dithering first.



#11 Jon Rista

Jon Rista

    ISS

  • *****
  • Posts: 25,208
  • Joined: 10 Jan 2014
  • Loc: Colorado

Posted 04 January 2020 - 06:04 PM

What does dithering have to do with the variation in pixel sensitivity, ie. FPN? Nothing. Thanks.

Wrong! 4.gif

 

It has everything to do with FPN. This is the reason we dither. If you stack a bunch of frames without any dithering or drift, the same pixels stack on top of each other, and the pattern ultimately shows through. If you dither but have drift, then the patterns slowly move in the direction of drift, they "correlate" through the stack, and cause correlated noise (walking noise, raining noise...streaked patterns in the signal). 

 

Dithering randomizes the position of the stars in each frame. Once aligned, the stars are in the same position, but the PATTERNS are now randomized. By dithering, you have effectively imparted a temporally random component to what would otherwise be a fixed pattern, and now, within the stack, it has become a temporally random noise. Therefor, it averages out like any other noise.

 

Flats can help mitigate per-pixel FPN variations, but they are not perfect, and there is usually some remnant pattern. So flats are not an ideal solution to FPN. Same goes for darks, they will usually match some pattern but not all. Further, CMOS cameras also exhibit things like RTS noise, where pixels may exhibit 2 or even 3 different discrete levels of intensity over time. Patterns can also change due to changes in gain, offset, and other things like USB Limit settings. So darks and flats are not the sole solution to FPN. In the long run, the pattern of noise in the master dark and master flat become a form of FPN themselves, and if you use short exposures (sometimes required with CMOS cameras) and stack LOTS of frames, then this can be another form of FPN that eventually exhibits.

 

Dithering solves these issues with FPN in the long run. It should not be used alone, as again it alone is insufficient, but combined with dark and flat calibration, dithering ensures that any remnant FPN is ultimately randomized through the stack allowing it to average out. If you had no FPN, then there would be no reason to dither. Not, at least, unless you were also planning to drizzle integrate. 


Edited by Jon Rista, 04 January 2020 - 06:08 PM.

  • okiedrifter, psandelle, TrustyChords and 1 other like this

#12 jhayes_tucson

jhayes_tucson

    Fly Me to the Moon

  • *****
  • Posts: 7,388
  • Joined: 26 Aug 2012
  • Loc: Bend, OR

Posted 04 January 2020 - 06:04 PM

Frank,

 

The fact that you can see the one dust mote indicates the flats are not well matched with the lights.  People worry about the exposure of flats, but as long as they aren't exposed too much and as long as you have enough exposures they should be ok - but only if they match the illumination of the lights.  I would try a clear sky or t-shirt flat and see if it works better.  If it is bright outside when you take the flats, make sure the imaging train is well shielded from light.

 

It's nice to see people talking more about pattern noise and dithering - but it's important to distinguish fixed pattern noise, which is identical in each frame, from PRNU due to variations in pixel response.  The latter is a form of pattern noise but the resulting noise in the image will depend on the scene being imaged - which makes it fundamentally different from FPN - even though there is a pattern to it.

 

Both FPN and PRNU would be helped by dithering - but the presence of the dust mote after calibration suggests the flats and calibration procedure are not optimal and should be addressed first.  

 

Frank

 

 

Frank,

FPN is due to PRNU and can be removed by flat fielding1,2,3.   PRNU is the acronym for "pixel response non-uniform" and it is a fixed characteristic of the image sensor itself.  FPN is due to a variation in the sensitivity of pixels across an imaging sensor so its effect is directly proportional to the irradiance of the incident light.  It is a signal modulator just like vignetting is a signal modulator.  Its name is descriptive of its effect on spatial "smoothness" across an image; but, it is no more a noise term than optical vignetting is a noise term.

 

John

 

 

(1) Janesick, James R. "Photon Transfer",SPIE Press, 2007.  pp 8, 9, 21, 30, 33, 88, 158.

 

2) Janesick, James R. "Scientific Charged-Coupled Devices, SPIE Press, 2000.  pp 318-332.

 

3) Crisp, Richard,  2018 SPIE ShortCourse Notes on CCD and CMOS imaging, 2018.


Edited by jhayes_tucson, 04 January 2020 - 06:05 PM.

  • Jon Rista and t-ara-fan like this

#13 Jon Rista

Jon Rista

    ISS

  • *****
  • Posts: 25,208
  • Joined: 10 Jan 2014
  • Loc: Colorado

Posted 04 January 2020 - 06:07 PM

Frank,

 

 

 

Frank,

FPN is due to PRNU and can be removed by flat fielding1,2,3.   PRNU is the acronym for "pixel response non-uniform" and it is a fixed characteristic of the image sensor itself.  FPN is due to a variation in the sensitivity of pixels across an imaging sensor so its effect is directly proportional to the irradiance of the incident light.  It is a signal modulator just like vignetting is a signal modulator.  Its name is descriptive of its effect on spatial "smoothness" across an image; but, it is no more a noise term than optical vignetting is a noise term.

 

John

 

 

(1) Janesick, James R. "Photon Transfer",SPIE Press, 2007.  pp 8, 9, 21, 30, 33, 88, 158.

 

2) Janesick, James R. "Scientific Charged-Coupled Devices, SPIE Press, 2000.  pp 318-332.

 

3) Crisp, Richard,  2018 SPIE ShortCourse Notes on CCD and CMOS imaging, 2018.

Aye, PRNU IS the variation in pixel response. ;) FPN arises as a result of PRNU.

 

To be fully clear, FPN can come from a gain/photon related component (PRNU), as well as a dark signal related component (DSNU)...the latter is often called DFPN to differentiate, but not all FPN comes from gain and photo response. 



#14 jhayes_tucson

jhayes_tucson

    Fly Me to the Moon

  • *****
  • Posts: 7,388
  • Joined: 26 Aug 2012
  • Loc: Bend, OR

Posted 04 January 2020 - 06:12 PM

Aye, PRNU IS the variation in pixel response. wink.gif FPN arises as a result of PRNU.

 

To be fully clear, FPN can come from a gain/photon related component (PRNU), as well as a dark signal related component (DSNU)...the latter is often called DFPN to differentiate, but not all FPN comes from gain and photo response. 

 

I provided clear reference to this.  FPN is due to variations in responsivity across an imaging sensor.   DFPN describes a spatial pattern in the dark current signal that arrises from a number of different sources and that's not the same as FPN.

 

John



#15 freestar8n

freestar8n

    Vendor - MetaGuide

  • *****
  • Vendors
  • Posts: 11,457
  • Joined: 12 Oct 2007
  • Loc: Melbourne, Australia

Posted 04 January 2020 - 06:29 PM

I provided clear reference to this.  FPN is due to variations in responsivity across an imaging sensor.   DFPN describes a spatial pattern in the dark current signal that arrises from a number of different sources and that's not the same as FPN.

 

John

I have provided multiple references to this.  I am just citing literature usage and common sense.  Pattern noise is of two main types: Fixed - and identical in all exposures - and response related, and dependent on the illumination as a multiplicative factor.

 

I prefer to adopt literature usage for clarity and to keep discussions aligned with how these things are described and in journals and textbooks.

 

It's a minor point here - but as people realize FPN is a legitimate noise term to be dealt with - I want to avoid a step backwards and lump PRNU in with it - since it is different and it is handled differently in professional writings.

 

The big dust mote visible isn't PRNU and it isn't pattern noise.  Its origin is dust somewhere in the system and the fact that it is visible after calibration means there is a problem in the calibration process - either the masters used or the math.  That is what needs to be addressed here.

 

Frank


  • Gleb1964 likes this

#16 freestar8n

freestar8n

    Vendor - MetaGuide

  • *****
  • Vendors
  • Posts: 11,457
  • Joined: 12 Oct 2007
  • Loc: Melbourne, Australia

Posted 04 January 2020 - 06:34 PM

Aye, PRNU IS the variation in pixel response. wink.gif FPN arises as a result of PRNU.

 

To be fully clear, FPN can come from a gain/photon related component (PRNU), as well as a dark signal related component (DSNU)...the latter is often called DFPN to differentiate, but not all FPN comes from gain and photo response. 

No - pattern noise arises as a result of PRNU - when multiplied by the irradiance.

 

FPN is just due to variations in bias offset and dark current.  Most texts specifically limit fixed pattern noise to noise that is present when there is no illumination.  That rules out PRNU - and I recommend sticking to that usage.

 

This isn't a convention I defined myself - I am going by the literature - and I think this distinction is useful.  Obviously others do also.

 

Frank


  • Gleb1964 likes this

#17 Jon Rista

Jon Rista

    ISS

  • *****
  • Posts: 25,208
  • Joined: 10 Jan 2014
  • Loc: Colorado

Posted 04 January 2020 - 06:43 PM

No - pattern noise arises as a result of PRNU - when multiplied by the irradiance.
 
FPN is just due to variations in bias offset and dark current.  Most texts specifically limit fixed pattern noise to noise that is present when there is no illumination.  That rules out PRNU - and I recommend sticking to that usage.
 
This isn't a convention I defined myself - I am going by the literature - and I think this distinction is useful.  Obviously others do also.
 
Frank

 
We are also going by literature, Frank. tongue2.gif Did you miss the references John quoted? I actually own Janesick's books. 
 

It's a minor point here - but as people realize FPN is a legitimate noise term to be dealt with - I want to avoid a step backwards and lump PRNU in with it - since it is different and it is handled differently in professional writings.

I've read enough literature (including all those John linked) and I agree with him here, PRNU is not a noise, PRNU stands for Photo Response Non-Uniformity, and is what GIVES RISE to gain FPN. PRNU itself is not a noise, though, in professional literature it describes the actual non-uniformity of the pixels. PRNU is part of the formula that gives you the FPN term (along with gain and irradiance, yes.) Professional literature also does often (usually) separate DFPN from FPN, to indicate the distinct natures of DARK fixed pattern noise that arises from DSNU (Dark Signal Non-Uniformity), from the fixed pattern noise that arises from gain and photons due to PRNU. That said, professional literature usually covers gain, offset and dark FPN, which all fall under the over-arching umbrella of "FPN" in general.
 
FPN is a legitimate noise term, but gain FPN is DUE TO PRNU, so I don't think we are "lumping" PRNU in as a separate term...it is part of the FPN term. Gain FPN is what we usually use flats to correct. Offset and Dark FPN are what we use darks to correct. Often, particularly in Janesick's works, Offset and Dark FPN are termed "DFPN". This is a convention that Crisp uses as well. So if you are ONLY referring to DFPN, then you should use the term that literature usually uses: DFPN. Often neither correction is absolutely perfect, hence the reason we dither.


Edited by Jon Rista, 04 January 2020 - 06:47 PM.


#18 jhayes_tucson

jhayes_tucson

    Fly Me to the Moon

  • *****
  • Posts: 7,388
  • Joined: 26 Aug 2012
  • Loc: Bend, OR

Posted 04 January 2020 - 07:17 PM

No - pattern noise arises as a result of PRNU - when multiplied by the irradiance.

 

FPN is just due to variations in bias offset and dark current.  Most texts specifically limit fixed pattern noise to noise that is present when there is no illumination.  That rules out PRNU - and I recommend sticking to that usage.

 

This isn't a convention I defined myself - I am going by the literature - and I think this distinction is useful.  Obviously others do also.

 

Frank

 

Frank,

Go back to the references that I cited.  You've thrown Janesick at me many times that maybe it's time for you to actually read and understand it.  FPN is not due to variation in bias offset and dark current.  FPN is due to pixel to pixel variations in responsivity across the sensor and it is signal dependent.  Can you give us a specific reference that limits the definition of fixed pattern noise to a noise that is present when there is no illumination?  Dark and bias current are signals that are spatially fixed across an imaging sensor with no illumination.  Are you referring to the noise generated by those signals?

 

John



#19 jhayes_tucson

jhayes_tucson

    Fly Me to the Moon

  • *****
  • Posts: 7,388
  • Joined: 26 Aug 2012
  • Loc: Bend, OR

Posted 04 January 2020 - 07:30 PM

 
We are also going by literature, Frank. tongue2.gif Did you miss the references John quoted? I actually own Janesick's books. 
 

I've read enough literature (including all those John linked) and I agree with him here, PRNU is not a noise, PRNU stands for Photo Response Non-Uniformity, and is what GIVES RISE to gain FPN. PRNU itself is not a noise, though, in professional literature it describes the actual non-uniformity of the pixels. PRNU is part of the formula that gives you the FPN term (along with gain and irradiance, yes.) Professional literature also does often (usually) separate DFPN from FPN, to indicate the distinct natures of DARK fixed pattern noise that arises from DSNU (Dark Signal Non-Uniformity), from the fixed pattern noise that arises from gain and photons due to PRNU. That said, professional literature usually covers gain, offset and dark FPN, which all fall under the over-arching umbrella of "FPN" in general.
 
FPN is a legitimate noise term, but gain FPN is DUE TO PRNU, so I don't think we are "lumping" PRNU in as a separate term...it is part of the FPN term. Gain FPN is what we usually use flats to correct. Offset and Dark FPN are what we use darks to correct. Often, particularly in Janesick's works, Offset and Dark FPN are termed "DFPN". This is a convention that Crisp uses as well. So if you are ONLY referring to DFPN, then you should use the term that literature usually uses: DFPN. Often neither correction is absolutely perfect, hence the reason we dither.

 

Jon,

I agree with what most of what you are saying; however, FPN is a signal modulator and although Janesick treats it as a noise term, that's only in the context of trying to characterize the spatial uniformity of a sensor.  FPN represents a systematic uncertainty in the measurement that can be arithmetically calibrated out of an image so it does not represent temporally uncorrelated noise (uncertainty)--in spite of the unfortunate fact that it contains the word "noise" in its name.  Of course the calibration process needed to remove the effects of FPN does add some measurement uncertainty to the result (as Janesick points out in chapter 8 , p111 of ref (1)).

 

John


  • psandelle likes this

#20 freestar8n

freestar8n

    Vendor - MetaGuide

  • *****
  • Vendors
  • Posts: 11,457
  • Joined: 12 Oct 2007
  • Loc: Melbourne, Australia

Posted 04 January 2020 - 07:33 PM

 
We are also going by literature, Frank. tongue2.gif Did you miss the references John quoted? I actually own Janesick's books. 
 

I've read enough literature (including all those John linked) and I agree with him here, PRNU is not a noise, PRNU stands for Photo Response Non-Uniformity, and is what GIVES RISE to gain FPN. PRNU itself is not a noise, though, in professional literature it describes the actual non-uniformity of the pixels. PRNU is part of the formula that gives you the FPN term (along with gain and irradiance, yes.) Professional literature also does often (usually) separate DFPN from FPN, to indicate the distinct natures of DARK fixed pattern noise that arises from DSNU (Dark Signal Non-Uniformity), from the fixed pattern noise that arises from gain and photons due to PRNU. That said, professional literature usually covers gain, offset and dark FPN, which all fall under the over-arching umbrella of "FPN" in general.
 
FPN is a legitimate noise term, but gain FPN is DUE TO PRNU, so I don't think we are "lumping" PRNU in as a separate term...it is part of the FPN term. Gain FPN is what we usually use flats to correct. Offset and Dark FPN are what we use darks to correct. Often, particularly in Janesick's works, Offset and Dark FPN are termed "DFPN". This is a convention that Crisp uses as well. So if you are ONLY referring to DFPN, then you should use the term that literature usually uses: DFPN. Often neither correction is absolutely perfect, hence the reason we dither.

Jon- It is interesting to be pointed to Janesick by you and John since I believe you both acquired Janesick on my recommendation in CN.

 

I agree with your description of PRNU as not a noise term - but I prefer to say does not give rise to FPN.  It gives rise only to pattern noise.  It is a minor point.

 

I don't have Janesick with me right now so I don't know what he specifically says - but it's possible he does consider PRNU to cause FPN.  If so - there are other writings that distinguish the two as I describe.  No matter what - it's important to be aware how different they are and how differently they impact an image.

 

Again - this is all great progress since at least I'm not being lectured to regarding signal vs. noise.  

 

Here is wikipedia lumping PRNU in as FPN - with the stipulation that the illumination is constant in all exposures:

 

https://en.wikipedia...d-pattern_noise

 

Here is an article cited over 1000 times that distinguishes the two as I describe:

 

https://ia.binghamto...hPDF/double.pdf

 

 

 

The two main components of the pattern noise are the fixed pattern noise (FPN) and the photo-response non-uniformity noise (PRNU) (see Fig. 1). The fixed pattern noise (FPN) is caused by dark currents. It primarily refers to pixel-to-pixel differences when the sensor array is not exposed to light.

There is a wide body of literature on using pattern noise in images to identify cameras - and it can be done from multiple exposures of entirely different scenes.  So it is more important for them to distinguish the two main types of pattern noise.

 

Here's another reference by someone well published in sensor literature:

 

https://ece.uwaterlo...r2008/Noise.pdf

 

 

 

Fixed Pattern Noise
• Fixed Pattern Noise is due to pixel-to-pixel variations in the absence of illumination
PRNU
• The issue of photo-response non-uniformity has not historically received much attention in the CMOS imager community
» although there is now some occasional mention of “gain nonuniformity”
• Like FPN, PRNU is essentially time-independent, but it is signal-dependent
• Both types of pattern noise can be specified in terms of either an rms or a peak-to-peak value, referenced to an average value

So - it's great you are using these terms and some literature may lump the two together.  But I am well read on this stuff and aware of possible different usages.  So I guess it isn't wrong to lump in the impact of PRNU with FPN - but there is good reason not to - and plenty of literature that makes it a point not to.

 

Frank


  • Gleb1964 likes this

#21 AXAF

AXAF

    Apollo

  • -----
  • topic starter
  • Posts: 1,050
  • Joined: 15 Sep 2004
  • Loc: MA

Posted 04 January 2020 - 07:49 PM

Wrong! 4.gif

 

It has everything to do with FPN. This is the reason we dither. If you stack a bunch of frames without any dithering or drift, the same pixels stack on top of each other, and the pattern ultimately shows through. If you dither but have drift, then the patterns slowly move in the direction of drift, they "correlate" through the stack, and cause correlated noise (walking noise, raining noise...streaked patterns in the signal). 

 

Dithering randomizes the position of the stars in each frame. Once aligned, the stars are in the same position, but the PATTERNS are now randomized. By dithering, you have effectively imparted a temporally random component to what would otherwise be a fixed pattern, and now, within the stack, it has become a temporally random noise. Therefor, it averages out like any other noise.

 

Flats can help mitigate per-pixel FPN variations, but they are not perfect, and there is usually some remnant pattern. So flats are not an ideal solution to FPN. Same goes for darks, they will usually match some pattern but not all. Further, CMOS cameras also exhibit things like RTS noise, where pixels may exhibit 2 or even 3 different discrete levels of intensity over time. Patterns can also change due to changes in gain, offset, and other things like USB Limit settings. So darks and flats are not the sole solution to FPN. In the long run, the pattern of noise in the master dark and master flat become a form of FPN themselves, and if you use short exposures (sometimes required with CMOS cameras) and stack LOTS of frames, then this can be another form of FPN that eventually exhibits.

 

Dithering solves these issues with FPN in the long run. It should not be used alone, as again it alone is insufficient, but combined with dark and flat calibration, dithering ensures that any remnant FPN is ultimately randomized through the stack allowing it to average out. If you had no FPN, then there would be no reason to dither. Not, at least, unless you were also planning to drizzle integrate. 

I am talking about addressing PRNU in single image frame, not in the aggregate of a final integrated product, where I agree that dithering, of course, reduces the spatial noise.



#22 jhayes_tucson

jhayes_tucson

    Fly Me to the Moon

  • *****
  • Posts: 7,388
  • Joined: 26 Aug 2012
  • Loc: Bend, OR

Posted 04 January 2020 - 08:18 PM

Frank,

I sure see a lot of stuff in that article that I disagree with but more than anything, it emphasizes why it's necessary to have a terminology that everyone can agree on.  Simply saying that something is in the literature doesn't necessarily make it "right"--particularly if every paper, article, website, or book uses the terminology differently.  For a long time, Janesick's "Photon Transfer" was treated as the go-to "Bible" for all things related to sensor characteristics around here--particularly by you.  And now, after repeatedly affirming the "professional nomenclature" used in that reference, you've tossed it out to meet your own purposes.  It's hard to tell what you are talking about when you change your definitions so easily.  I'm happy to give you the benefit of the doubt that you know what you are talking about; but, it gets hard when you shift directions with every thread.  I provided a correction, I gave you the definitions and I gave you the specific references--from a source that you yourself affirmed is correct and you still turned around and ignored what Jon and I are telling you.  This kind of behavior seriously erodes your credibility and it makes me doubt your understanding of this stuff at the most basic level.  In my view, that's why it is particularly important to understand the differences between signal and noise right up front.  Get that screwed up and the whole discussion veers into the weeds almost from the get-go and that's why I've mostly given up arguing with you.  You appear to be more interested in being the expert than in being right...and that's hard to deal with.

 

John


  • Jon Rista, johnpane and Terry White like this

#23 freestar8n

freestar8n

    Vendor - MetaGuide

  • *****
  • Vendors
  • Posts: 11,457
  • Joined: 12 Oct 2007
  • Loc: Melbourne, Australia

Posted 04 January 2020 - 09:04 PM

John - if you have a scene and you dither it, then the FPN in each exposure will be the same - but the pattern noise will be different because the illumination has shifted across the sensor.  It behaves completely differently.  That is why much of the literature - but not all - makes this distinction.

 

In the case of astronomical imaging with dithering - and in particular with regard to this thread - I think it's important to distinguish the two.  Then we can talk about one noise term that is constant in each dithered image - and another that is not constant in each dithered image.  Though both are pattern noise terms.

 

I have been motivating the concept of pattern noise in CN over the years and it was hard to get the basic idea across.  Now I am hoping to take it an additional level.  And that extra level will I hope make sense to people now that the basic ideas of FPN have been accepted - and are helping understand dithering - etc.

 

In short - there is less motivation to distinguish PRNU from FPN when the scene is either uniform or it does not change.  But when you dither, the noise in the image caused by PRNU will change - while the FPN does not.  That is the context here - and it is the context in the literature where that distinction is made.

 

People often talk about noise in their images and are concerned about PRNU causing noise in the dark regions - where PRNU is likely not involved at all.  It could be if the sky background is really bright - but it's unlikely.  PRNU would only reveal itself in smooth, bright parts of the image. 

 

Frank


  • Gleb1964 likes this

#24 jhayes_tucson

jhayes_tucson

    Fly Me to the Moon

  • *****
  • Posts: 7,388
  • Joined: 26 Aug 2012
  • Loc: Bend, OR

Posted 04 January 2020 - 09:26 PM

John - if you have a scene and you dither it, then the FPN in each exposure will be the same - but the pattern noise will be different because the illumination has shifted across the sensor.  It behaves completely differently.  That is why much of the literature - but not all - makes this distinction.

 

In the case of astronomical imaging with dithering - and in particular with regard to this thread - I think it's important to distinguish the two.  Then we can talk about one noise term that is constant in each dithered image - and another that is not constant in each dithered image.  Though both are pattern noise terms.

 

I have been motivating the concept of pattern noise in CN over the years and it was hard to get the basic idea across.  Now I am hoping to take it an additional level.  And that extra level will I hope make sense to people now that the basic ideas of FPN have been accepted - and are helping understand dithering - etc.

 

In short - there is less motivation to distinguish PRNU from FPN when the scene is either uniform or it does not change.  But when you dither, the noise in the image caused by PRNU will change - while the FPN does not.  That is the context here - and it is the context in the literature where that distinction is made.

 

People often talk about noise in their images and are concerned about PRNU causing noise in the dark regions - where PRNU is likely not involved at all.  It could be if the sky background is really bright - but it's unlikely.  PRNU would only reveal itself in smooth, bright parts of the image. 

 

Frank

 

 

You are not making sense and that may be why you are having such a hard time of getting "the basic idea" across.  Please read Janesick (ref 1 above.)  FPN is caused by PRNU, it is signal dependent and it is not a noise term.  PRNU is a characteristic of the detector and it is a signal modulator that can be calibrated out by flat fielding.  Dithering helps to remove any residual error in calibration by spatially averaging the residual noise terms introduced by the flat fielding process.  The dithering distance has to be larger than the autocorrelation distance of the FPN and there has to be a sufficiently large number of frames to drive the uncertainty to a sufficiently low level that it's not noticeable (N>33 reduces the uncertainty by more than 87%.)  It is true that FPN is most noticeable in the bright regions of an image.

 

John


  • Jon Rista likes this

#25 freestar8n

freestar8n

    Vendor - MetaGuide

  • *****
  • Vendors
  • Posts: 11,457
  • Joined: 12 Oct 2007
  • Loc: Melbourne, Australia

Posted 04 January 2020 - 09:46 PM

You are not making sense and that may be why you are having such a hard time of getting "the basic idea" across.  Please read Janesick (ref 1 above.)  FPN is caused by PRNU, it is signal dependent and it is not a noise term.  PRNU is a characteristic of the detector and it is a signal modulator that can be calibrated out by flat fielding.  Dithering helps to remove any residual error in calibration by spatially averaging the residual noise terms introduced by the flat fielding process.  The dithering distance has to be larger than the autocorrelation distance of the FPN and there has to be a sufficiently large number of frames to drive the uncertainty to a sufficiently low level that it's not noticeable (N>33 reduces the uncertainty by more than 87%.)  It is true that FPN is most noticeable in the bright regions of an image.

 

John

You are not reading what I have written - and have not explored the literature.

 

PRNU causes pattern noise in the image - but that noise isn't constant in each exposure like FPN.

 

PRNU is not "noise in the image" and it doesn't even cause "pattern noise that is constant in each image."

 

PRNU combined with illumination produces noise in a given image - and the noise in the image will depend on the illumination.  PRNU results in a form of pattern noise specific to the nature of the illumination in each exposure - and that is important to distinguish from FPN - which is fixed and constant in each exposure - independent of the illumination.

 

I am familiar with how dithering improves pattern noise and described it and its mathematics many times in CN.  The point I am making here is that dithering helps with all pattern noise - but it won't help with PRNU in dark parts of the image at all.  Because there is minimal or no pattern noise related to PRNU in dark parts of the image.  But there is always FPN.

 

Everything I am describing is discussed in the literature - though some literature focused on static scenes or fixed illumination will not make this distinction - because it isn't pertinent.  But it is quite pertinent here.

 

For the OP - your image is highly stretched and appears to show noise in the dark regions - and it also shows evidence of a problem with dust not being calibrated away.  The problem with the dust indicates an issue with calibration - and that's the main thing to address.  The noise in the dark parts of the image is likely not related to PRNU and instead is mostly due to pattern noise in the dark - and dithering should help with that.  But dithering won't help with that dust spot - and the spot shouldn't be visible in the first place.

 

Frank




CNers have asked about a donation box for Cloudy Nights over the years, so here you go. Donation is not required by any means, so please enjoy your stay.


Recent Topics






Cloudy Nights LLC
Cloudy Nights Sponsor: Astronomics