Printable Version of Topic

Click here to view this topic in its original format

Unmanned Spaceflight.com _ Juno _ Discussion of stray light in Juno Earth flyby images

Posted by: Gerald Aug 27 2015, 10:02 PM

This is now a first step of a somewhat sophisticated analysis of the ghosts, the success of which is to find out.
The ghosts are best feasible outside the bright Earth. So I've masked all efb12 subframes to just show the ghosts in front of dark background.
Example:


This is the average of all these masked subframes, constraint to the non-black pixels:

This average shows remnants of the source images. Most of this can be averaged away by decomposing the image in a horizontal and a vertical mean brightness function.
For averaging, the brightness values are temporarily gamma-corrected with gamma = 2, considering the square-root encoding of the raw image.
The two functions can then be composed to result in this mean ghost image:

This image is intended to serve as a 0th approximation of a flatfield for the ghosts.


Posted by: Gerald Aug 28 2015, 09:35 PM

After horizontal point noise filtering of the 0th "ghost flatfield",


and application to the masked ghost images, e.g. with this result,

8-fold vertical condensing of 0th-flatfielded ghosts

reveals three types of ghosts which can easily be distinguished:
- a sharp ghost below the bright object,
- a blurred ghost below the bright object,

- and a vertically close to structureless ghost above the respective bright object.


The ghosts below an object appear to be decomposed into five columns.
Columns 1, 3, and 5 contain the sharp ghost,
columns 2 and 4 contain the blurred ghost.
The ghost above a bright object isn't obviously decomposed into columns.

The 8-fold vertical compressed sharp ghosts of EFB12 are sufficiently clearly structured, that they can be identified with features on Earth.

The blurred ghosts are sufficiently sharp to show that they originate in vertically displaced strips of the source object relative to the sharp ghosts.

The three thus far identified types of ghosts are to be investigated separately in more detail.

Posted by: Gerald Aug 29 2015, 06:12 PM

Subframes 0 and 81 show the same part of Earth, with an inaccuracy of about 20 pixels near the center of the subframes:


That's roughly corresponding to a full 360° rotation of Junocam, and an offset due to the relative motion of the spacecraft.
(Nominally a rotation takes about 30s / 0.38s = 79 subframes. Based on this, the offset is about 2 subframes of the 81 subframes difference.)

The ghost at the left in subframe 71

shows features visible in the green framelet of subframe 76, and it almost fills in part of the masked strip between the blue and the green filter in subframe 75.

That's near the presumed vertical optical center. The horizontal offset of the ghost in this example is about 32 pixels to the right.
As a first estimate, the origin of the sharp ghost is vertically 4 subframes off the center, or about 4 * 360°/81 = 18° (considering the average relative motion of spacecraft).

Same exercise with subframes 67 (ghost)

and 71, offset of the ghost about 12 pixels to the left:


And just to be sure, a crosscheck with subframes 66 (ghost)

and 70, again offset of the ghost about 12 pixels to the left:


So, the sharp ghosts seem to be distorted horizontally. This doesn't fit well to the small change of perspective by a difference of just 4 subframes.
Optical distortions (pinhole / barrel) would be of a similar order of magnitude as the observed distortions.

This step provides a basis to decide, whether the analysis of the ghosts can be continued with 2d methods allone, or whether map projections and 3d simulations become necessary at this point.
Now, continuing merely with 2d methods appears feasible, with the advantage, that less a-priori knowledge about the target is needed.

Posted by: Gerald Aug 31 2015, 10:41 AM

An idea of the draft pinhole simulation of the geometry of the sharp ghosts, derived from the raw color version of a merged efb12:


It needs further geometric refinement (by about an accuracy factor of 5 to 10), e.g. by considering the Brownian K1 parameter, to be useful for cleaning.
These two gifs may provide an idea of the remaining distortion, first the bright portion of the image:

The (more subtle) sharp ghost portion (best visible in the leftmost of the five columns) :

Posted by: Gerald Sep 2 2015, 03:21 PM

It took me some time to improve the geometric distortion of the images needed as a starting point for the simulation of the sharp ghosts.

Another ingredient is a ghost-type specific "flatfield". It's summarized in the following gif.
The idea is specific averaging of the "flatfield" from the sharp ghosts, followed by manual selection and copying of the central column to the first and fifth, since it appears to be of the highest quality.
Intermediate steps are averaging of the approriate subframes, decomposition in horizontal and vertical average with consecutive composition, horizontal point noise filtering, clipping of the two columns with the blurred ghosts, manual selection of the highest quality column as a prototype:


The next gif shows steps to narrow down the simulation of the sharp ghosts by using the "flatfield" as an intensity mask; the first frame is a distorted crop of the raw color merged efb12; the last frame is a subframe made of three raw framelets.
The gif shows steps to (approximately) derive the sharp ghost in the raw framelets from a distorted crop of the merged efb12:

Posted by: Gerald Sep 3 2015, 11:48 PM

12 ghost images, cropped from the raw subframes (#64 - #75), flatfield-enhanced with the 0th "ghost flatfield", 8-fold vertically condensed, and composed to an overview image:


As a comparison: Ghosts, simulated from the raw color merged EFB12, based on the thus far achieved modeling of the "sharp" ghosts, also 8-fold vertically condensed:

The geometry is roughly ok, it may however need some further refinement. Two not yet modeled properties of the original ghosts are going to emerge:
- The sharp ghosts seem to be superposed with other ghosts or scattered light.
- The sharp ghosts aren't perfectly sharp, but blurred a bit.

The bright vertical lines in the enhanced original ghosts are probably due to hot pixels, to be investigated separately.

Posted by: Gerald Sep 4 2015, 08:25 PM

A first estimate of "how" blurred the ghosts are.
To get an idea, I've blurred the raw merged EFB12 with Gauss blurs of exponentially increasing radii starting with radius 1 pixel with multiplicative steps of about sqrt(2), more formally radii sigma_n = sqrt(2)^n, for n = 0,...,10, the largest radius being 32 pixels.
The relative weight for the blur is exp(-r^2/(2 sigma^2)), with r the pixel distance to the central pixel; the blur is calculated over a square from -(3 sigma + 1) to +(3 sigma + 1), decomposed into two consecutive blurs, a horizontal and a vertical one.
As color values I've used the linearized values, which have been square root encoded again, after applying the blur.

Some of the resulting blurred EFB12 images contain insets of cropped and 8-fold vertically condensed raw subframes. Here two excerpts of cropped versions as animated gifs:



For the "sharp" ghosts a sigma of 1.4 pixels looked similar, for the "blurred" ghosts a sigma of 22.6 pixels. The values for sigma should be ok within a factor of two as an error estimate.


Posted by: Gerald Sep 6 2015, 02:17 AM

The source of the blurred ghosts seems to be about 0.44° less displaced than the source of the sharp ghosts, hence about 17.56°, preliminarily assuming the source of the sharp ghosts being displaced by 18.00°.

Here a separate simulation of the sharp ghosts (Gauss blur with a sigma of 1.4 pixels), the blurred ghosts (Gauss blur with a sigma of 22.6 pixels), and the merged efb12 in raw colors.
I didn't apply intensity filters for the ghosts in this simulation. Images are cropped to the upper (southern) fragment.
I'm posting the three individual images as JPGs instead of an animated gif to avoid considerable loss of quality.


The three images may look displaced at first glance, but they are aligned up to the accuracy of the simulation, a few pixels for most parts of the images.

The comparison shows, to which extent the simulation applies. Particularly the overlap will require additional considerations.
But before, the third identified type of ghosts can be investigated in more detail, at least, and some refined intensity analysis should be possible.

Posted by: Gerald Sep 30 2015, 10:10 PM

Some regions of the efb images can be assumed to be colored almost exclusively by ghosts.
If these ghosts can be assumed to be known except their pixel-wise weight, the weight being the same for all framelets at a given CCD pixel position, the weights can be calcuated in most cases, provided a sufficient number of subframes are available.

After applying this method, I'm rather sure, that in addition to the sharp and the blurred ghost below an object, other types of ghosts below an object need to be assumed.

From right to left in this visualized equation are a column vector of 2 raw subframes, a 2x2 matrix of simulated ghosts with constant weight 1, and a column vector of the calculated weights:


You may note, that the column (at the left) with the weights looks rather heterogenious, which shouldn't be the case for the correct weights.

The next obvious step is hence to narrow down the remainig ghosts / stray light, such that the calculated weights get reasonable.

Here an article summarizing some of the theoretical background:
 junocam01_some_related_algebra.pdf ( 105.77K ) : 707

Posted by: Gerald Oct 9 2015, 03:19 AM

A preliminary analysis of EFB03 hints towards the structure of the target object being relevant for the poorly structured stray light.
 junocam02_about_efb03_straylight.pdf ( 117.54K ) : 923

The test tries to determine the degree of linear dependence of framelets showing mostly stray light.
Linear dependent framelets would indicate few relevance of the structure of the target object.
EFB03 is particularly well-suited for this purpose.

According matrices of correlation diagrams, mentioned in the above article, cases 1x1, 2x2, and 3x3:


Clear diagonal lines in these diagrams would point towards the structure of the target object being not relevant for the stray light.

Posted by: Gerald Oct 24 2015, 12:14 PM

A fist step towards an empirical description of the stray light in EFB03:

Decomposing the framelets respectively into three appropriate horizontal substripes allows linear regression for these substripes, if constrained to averages within vertical stripes.
The regression parameters can then be approximated by Gauss functions.
Graphical synopsis:



An animated gif of six images, indicating the dependence of the parameters of the linear regression fragments on the selected vertical stripe:


More detailed technical description:
 junocam03_empirical_efb03_straylight_I.pdf ( 195.17K ) : 573


Next steps will probably be an attempt to refine the 1-dimensional approximations, and to extend the approximations to a closed 2-dimensional description over the whole image.

Posted by: Gerald Dec 22 2015, 01:52 PM

Evaluating a simple function at a small number of points is rather easy.
Finding a simple function, which fits well to a time series, however, can be hard.
So this article is a moderately tough one:
 junocam04_empirical_efb03_straylight_II.pdf ( 263.87K ) : 771

It's about finding a family of functions able to fit into a 1-dimensional time series of grey values, as derived from EFB03.
The functions are designed to overcome some of the weaknesses of https://en.wikipedia.org/wiki/Gaussian_function, especially to provide https://en.wikipedia.org/wiki/Periodic_function, https://en.wikipedia.org/wiki/Skewness, and https://en.wikipedia.org/wiki/Kurtosis.

I hope, I'll find time the next few days, to create several diagrams visualizing some of the methods and data.

Posted by: Gerald Dec 24 2015, 09:04 PM

This graphics visualizes some properties of a family of peaks:



These diagrams show the horizontal variability in EFB03, of properties similar to the above:


Posted by: Gerald Dec 29 2015, 06:12 PM

Possible physical root cause for sharp ghosts in Junocam images:
EFB12 subframe compared with camera CCD:


Path of light ray including possible reflection, z-axis (optical axis) as vertical:

Both attachments use distorted and cropped versions of figure 12 of http://link.springer.com/article/10.1007/s11214-014-0079-x?sa_campaign=email/event/articleAuthor/onlineFirst.

Posted by: Gerald Jan 4 2016, 06:14 PM

Some more explanations regarding the above article:

As described in the article, I've applied methods related to the Newton method to fit peaks into the EFB03 data. The idea is, to find a local minimum of the square error sum. For differentiable functions, the derivative is zero at the minimum of a function. The Newton method can find such a point efficiently, if the derivative is locally sufficiently similar to a linear function. Unfortunately, the error functions defined immediately by the RMS error of a ("power-law") peak with respect to the actual data isn't that well-behaved at the guessed starting points of the iteration. But by using a sufficiently high power of the RMS error, the resulting function can be made sufficiently well-behaved. The following graphics visualizes the underlying principle for the 1-dimensional case:


PDF version:  NewtonMethod_and_Modification.pdf ( 118.43K ) : 869


An explanation of zeta has been pending:
Here two attempts to give the parameter zeta an intuitive meaning, visually (Fourier series and effect on peak)

and audible, with each "beat" starting a new damping by increasing zeta in a linear way. Parameter u1 varies with each "beat" in a total of two cycles. The audio version samples only the Fourier series, not the respective derived peak. (The file is packed twice with 7-zip, for effective compression, and to obtain a .zip extension.)
 ZetaSamples1.0_2.0_0.2_1.01_1.41_0.004_400_400_2.wav.7z.zip ( 269.26K ) : 353


And back to EFB03:
This graphics shows linear regression data ("raw") of vertical stripes of width 100 pixels obtained from near the left and from near the right side of EFB03,
the best-fit "power-law" peaks with zeta=1, and the residuals (remaining errors).


There appear to be systematic errors with respect to the considered family of peaks. This effect is more distinct near the right of EFB03 than near the left side. I'm presuming, that these deviations can be reduced a little, but not considerably by including zeta as variable peak parameter. Inferring zeta is rather fragile and probably time-consuming, therefore I'm looking for other options first.

My favored approach will be a description by a sum of two peaks, a narrow high and a wide shallow one. Since the horizontal variability of the peak parameters clearly cannot be described by a single peak, I'm considering to investigate multi-peak approaches as the next step to narrow down the 2d-structure of EFB03 stray light.

The additional math needed to feed into some modified Newton method doesn't look difficult at first glance. Although I don't know yet, how well-behaved the approach will be numerically. I guess, that dedicating another about two weeks will return first (continuous) 2d approximates.

Other related tasks appear at the horizon, e.g.
- writing a ray tracer to explain/model the 2d structure of the stray light physically, and
- pinning down the effect of TDI regarding integration over some neighboring color filter (probably responsible for horizontal substripes within framelets).

Posted by: mcaplinger Jan 4 2016, 07:20 PM

QUOTE (Gerald @ Jan 4 2016, 10:14 AM) *
As described in the article, I've applied methods related to the Newton method to fit peaks into the EFB03 data.

Keep in mind that EFB03 was taken in a mode that is unlikely to be duplicated for Jupiter observations (lots of TDI in a visible channel). We will probably only use a lot of TDI for the CH4 channel.

In the diagram you show above of the CFA, be aware that there is a light shield between the top of the CCD and the back of the optics with an aperture nominally 0.231x0.473 inches centered on the center of the sensor. This is intended to block stray light paths from the sensor bond wires and metallization on the sensor package, though there could be some misalignment and some small paths could still exist.

Mods: it might make sense to create a subforum for this material, much as was done with the MSL discussion of technical details about the cameras.

Posted by: Gerald Jan 5 2016, 01:24 PM

Thanks a lot, that's helpful information.
Provided the light shield is centered horizontally, but displaced vertically, such that the CH4 filter and the red filter are fully illuminated, this would explain the superposition of two ghosts below the primary image of the target object, and the presence of only one ghost above the primary image.
The three bright rectangular areas next to the CH4 filter may add the sharp ghost below the primary image. In contrast, the corresponding bright rectangular areas near the red filters are covered completely by the light shield.
The light shield itself might add the blurred ghost below and the ghost above the primary image. Thus far I found out, that (one of) the reflecting feature(s) adding the ghost above the primary image should be close to the lower end of the red filter.

The blue filter appears to be shadowed from some stray light in a zone next to the CH4 filter. But I couldn't yet decypher, whether that's a side effect by design to protect the CH4 filter from stray light, or whether the CH4 filter is thicker than the blue filter, and casts a shadow.

There is a large number of possibly relevant detail about the camera, but probably only a small fraction of which will turn out to be actually relevant for the calibration of the images. I presume, that a detailed plan of the geometry of all surfaces possibly in contact with non-negligible light eventually reaching the sensors together with BDRF data (including anti-reflective coatings), and the refractive indices of the translucent materials would be helpful. But I don't expect this detail being readily available or cleared for publishing. Might be, you could provide the thickness of the color filters, and the z-position (relative to the CCD) and thickness of the light shield, together with the geometry of the chamber between the CCD and the optics.
In the meanwhile, I'll work with the publicly available documents and images to narrow down the relevant detail.

Posted by: mcaplinger Jan 5 2016, 04:27 PM

QUOTE (Gerald @ Jan 5 2016, 05:24 AM) *
a detailed plan of the geometry of all surfaces possibly in contact with non-negligible light eventually reaching the sensors together with BDRF data (including anti-reflective coatings), and the refractive indices of the translucent materials would be helpful.

I'll see what we can release. Some of that doesn't even exist -- for example, no BRDF measurements were made for this non-radiometric instrument. The ray trace of the optics in the Junocam paper is accurate if non-quantitative.

BTW, the most rigorous attempt I'm aware of to characterize the radiometric properties of a pushframe system (including stray light) can be found in "Inflight Calibration of the Lunar Reconnaissance Orbiter Camera Wide Angle Camera" http://asu.pure.elsevier.com/en/publications/inflight-calibration-of-the-lunar-reconnaissance-orbiter-camera-w (not open access, unfortunately.)

Posted by: Gerald Jan 5 2016, 06:35 PM

Thanks, great! Good to know, that the ray trace in the paper is accurate. I've been considering it to determine the refractive index and geometry of each of the lenses, if explicite data won't be published. So I know by now, that this approach will make sense.
Looking forward to data about the interior of the housing you can release. Approximate data might help already as initial values for approximation methods.
I'll consider to buy the LROC paper, in case I'll run out of ideas.

Posted by: mcaplinger Jan 6 2016, 11:11 PM

Here's a dimensioned drawing of the Junocam color filter array (dimensions in mm). In looking at EFB12, I would venture to guess that a lot of the stray light features are actually interline smear from the clear areas in the CFA adjacent to the CH4 region, not all of which are occluded by the light shield. These are exacerbated by the short exposure times we had to use in EFB.

Reflections off the light shield are not very plausible, as the shield and all of the internals of the optics are bead-blasted and black anodized. The AR coatings on the glass surfaces are as good as we could obtain but certainly there are paths from the filter edges, from the CCD die, bond pads, metallization, etc.

Since the instrument is not intended to be radiometrically precise and the stray light is only especially visible off the limb, I'm not thinking this is going to be much of an issue for most applications. I'm still more concerned about band-to-band registration, for which no perfect model yet exists.

Mods: again, I suggest this material be moved to a subforum as it's unlikely to be very interesting to most.

[moderator note: A Juno subforum will probably be created soon and the Juno thread split and/or reorganized when this happens]




Posted by: Gerald Jan 7 2016, 01:50 PM

Thanks for the technical plan. This clarifies several of the observed effects, e.g. the narrow horizontal substripes in EFB03 due to the small gap between the color stripes.

QUOTE (mcaplinger @ Jan 7 2016, 12:11 AM) *
I'm not thinking this is going to be much of an issue for most applications.

Here a 16x enhanced crop of EFB01, probably showing the sharp ghost of our Moon:

Since it's TDI 1, I've been interpreting the elongated shape as reflected light rather than smear (with some uncertainty).
It seems, the light of the moon happened to hit only one (causing the sharp ghosts) of at least three areas causing ghosts.
I'm expecting this to cause issues mainly in cases, when a bright target is displaced about 18 degrees (vertically) relative to a dark target of interest, think e.g. at attempts to observe auroras, or at areas which are dark in one spectral band.
QUOTE
... there may be some imaging during approach and earlier on the first orbit ...

Some images similar to the Moon image during approach addressing specifically the possibly reflecting or smearing areas would certainly help calibration.

QUOTE (mcaplinger @ Jan 7 2016, 12:11 AM) *
I'm still more concerned about band-to-band registration, for which no perfect model yet exists.

Pinning this down to subpixel precision looked much easier to me than quantifying the fuzzy stray light and smear, so I was inclined to do the difficult things first.
But just to be sure, I'll be going to shift priorities towards image geometry, since this needs to be elaborated anyway, including the geometry of the ghosts.

Powered by Invision Power Board (http://www.invisionboard.com)
© Invision Power Services (http://www.invisionpower.com)