[X] My Assistant

[Bjorn Jonsson]

This may be a bit specialized to say the least but if anyone can help I'd be grateful. I have now downloaded most of the raw Cassini images from the PDS imaging node plus the coiss_0011 volume from the PDS rings node. Volume coiss_0011 contains calibration stuff like flatfields, masks for removing dust rings and mottling etc.

I have been implementing calibration code which will eventually be added to my IMG2PNG utility, assuming there is interest in this. This utility is mentioned in this thread: http://www.unmannedspaceflight.com/index.p...wtopic=291&st=0 .

My code has been largely based on the CISSCAL source code on volume coiss_0011 but cleaned up a bit and rewritten using C++. This has been very successful despite a wide variety of rather arcane data formats on the coiss_0011 volume (bigendians, littleendians, IEEE and VAX floating point numbers, a weird TIFF file containing floating point data etc.). For example, flatfielding, linearization etc. works very well and in particular, for images of low contrast targets like Saturn and Titan makes a big difference if the images are to be greatly sharpened (no mottling or dustrings).

There is one problem though: Darkcurrent subtraction. This is a problem, the CISSCAL code for this is somewhat complex and may contain some bugs - what it does is to generate 'synthetic' darkcurrent images based on a model taking various factors into account. These 'synthetic' darkcurrent images get written to disk so each of them only has to be generated once. If anyone has 'sample' synthetic darkcurrent images I would greatly appreciate if at least one of these could be emailed to me since without this it is almost impossible for me to know if the images my code is generating are correct.

These images should (assuming my code is correct !) have names like \calib\darkcurrent\darks04228\n_dark027102410100_04228.img where '\calib' is in my case the calib-directory on the coiss_0011 DVD volume (which is lying on my hard disk).

Bjorn

[ugordan]

Can anyone help me on the whole calibration/radiometric correction topic? Is there a guide on the net on how one creates "true" color images from RGB filters (or those similar to them, e.g. in Cassini's case R -> CB1)? I'm basically a n00b when it comes to image processing at the level of calculating radiances and stuff. I'm well aware that the images the spacecraft takes are usually setup so that each exposure maximizes the S/N for that filter combination so, obviously, stacking three channels together doesn't produce the right results.

For starters, I don't want to do all the fancy stuff like flatfielding, de-banding, dark current removal etc, but I do 8to12 bit conversion when necessary as well as bias subtraction.

I've come across a table at the end of Cassini ISS description pdf, it lists values called omega for a given filter combo for both cameras and it's supposed to describe the amount of light each filter passes through. So I've tried compositing images by normalizing exposure_duration * omega for all three filters. The results actually look pretty good when I use this with Titan WAC RGB/RGV frames, I get the same colors as public ISS images turn out. With the NAC, however, things are different - I always end up with an underexposed blue channel and the whole image becomes too orange-reddish. I don't know if there's any more corrections I am missing like quantum efficiencies of the CCD or whether they're already incorporated into omega values.

Like I said, I'm not aiming for very scientifically accurate images, only those that I can say look close to the real thing. Any help would be appreciated...

[Bjorn Jonsson]

I have written a program that calibrates Cassini images, including flatfielding which is very important for low-contrast targets like Saturn and Titan. I will probably make it available soon - I have been too swamped in Cassini images recently to do so.

An underexposed blue channel is a familiar problem. Also in many cases it is necessary to use synthetic color, e.g. synthetic green from CB1 and blue or synthetic CB1 from CB2 and green (or even blue).

[ugordan]

I always thought the following press image was too weird:

It's said to be taken through a RGB combination, but the colors look all but normal. Even if I try to stretch the blue channel in my composites, that still gives me a weird blue fringe in other areas which are naturally bluer (atmospheric limb, etc.). It's as though some of the filters have a gamma function applied when I compare my images to the CICLOPS ones.
As for the blue channel underexposure, I don't really understand it - it should cancel out by a much longer exposure that the BL1 images are usually taken, but it doesn't.

Does it really come down to mixing various filters (in a sense that the R channel in Photoshop is not just RED/CL2 but a mix of both RED and GRN filters?) in sRGB colorspace to get a "true" color image? I'm lost...

[Bjorn Jonsson]

Actually "underexposed" is not what's happening in the blue filtered images - when calibrating the final result looks identical regardless of exposure time (at least in theory).

As for synthetic color, yes you do something like R=a x CB2 + (1-a) x G when using CB2 and green to get synthetic red.

[ugordan]

Actually "underexposed" is not what's happening in the blue filtered images - when calibrating the final result looks identical regardless of exposure time (at least in theory).

That would be my reasoning also. I gather the only difference with varying exposures would be in the final calibrated image S/N ratio and whether there would be overexposed areas in the image.
I still don't understand why, then, the BL1 channel turns out too dark when I do the above normalization. Is there an additional factor that needs to be taken into account? Perhaps a correction for the bandpass of the filter combo? IIRC, one of the blue NAC filters has a pretty narrower bandpass than the other.
Is there actually ANY deterministic way to produce a true color image from the RED/GRN/BL1(BL2) images or is it in the end just guesswork and finding out what looks right?

[ugordan]

I've been experimenting with "deterministic" ways to produce true color images. I've also been playing around with procedures to calibrate the images, adding flatfielding to my rudimentary converter.
My problem is, using NAC RED/GRN/BL1 filters and integrating the trasmittance with QE and solar spectrum, I wind up with composite images that tend to underexpose the blue channel. If I try correcting for the Sun's spectrum, I get results which, for Saturn, give more "natural" looking images, but make the icy moons look too blue.

I've attached two sample images (they weren't taken at the same time, but were both taken through CB1/GRN/BL1 filters)

The left image is the result when the formula cisscal uses is applied. Also it's flatfield-corrected. The right one is using a correction for the fact the Sun emits less in the blue end and it's not corrected for the flatfield (note the dust ring on the rings).

Which of the two would you say looks more like true color?

[Bjorn Jonsson]

I've attached two sample images (they weren't taken at the same time, but were both taken through CB1/GRN/BL1 filters)...

Which of the two would you say looks more like true color?

The right one definitely looks more realistic but based on what I've seen through telescopes I'd say the color is too saturated in both cases.

BTW now is a good moment to 'announce' a new version of my IMG2PNG utility:

http://www.mmedia.is/bjj/utils/img2png/

Previous versions were announced/discussed in this thread:

http://www.unmannedspaceflight.com/index.php?showtopic=291

The major change in this new version is that it can calibrate Cassini images. It should be noted that this is not a user friendly utility, neither the 'installation process' nor the user interface. For example it runs in a command prompt window so if you are not comfortable with that don't use it ;-). I plan to release a version with an improved user interface later but it's not a high priority.

Also you will need to download lots of data from the COISS_0011 Cassini DVD volume to be able to calibrate Cassini images.

Any feedback is welcome.

PS I'm attaching an image showing the importance of flatfielding when working with low contrast images. The two images are processed in a similar way (including sharpening with an unsharp mask) with the main difference being that the left one was not flatfielded. This is a section of a BL1 image of Saturn.

Attached thumbnail(s)

 

[tfisher]

BTW now is a good moment to 'announce' a new version of my IMG2PNG utility:

I am trying to use IMG2PNG to process Titan images, with mixed success. Maybe you can give me some help.

First of all, I have been trying to get away without downloading the entire calibration DVD. Instead, I've only downloaded the files named when img2png pops up a dialog complaining it didn't find them. Maybe you can tell me already that I'm doomed with this strategy.

For some images, it doesn't seem to try to do calibration at all. For instance, I download this IMG file and the corresponding LBL file. But img2png -r applied to this produces a file with only 5 distinct shades of gray, and no complaints about missing calibration files. Adding -s10 or somesuch doesn't increase the number of shades.

On the other hand, for an image like IMG/LBL the program requests several calibration files, and after they are present produces an output with 71 distinct shades of gray and partial flatfield correction. (From a 12bpp imager, I would have hoped for more than 71 shades, but at least this result resembles the correct thing!)

Am I doing something wrong?

[Bjorn Jonsson]

Am I doing something wrong?

Probably not - I discovered that in a few cases IMG2PNG didn't complain if it didn't find a calibration file so you probably need to download a few additional calibration files.

There is a new IMG2PNG version available that should always complain when calibration files are missing:

http://www.mmedia.is/bjj/utils/img2png/img2png_exe.zip

This file only contains a new version of the executable so it is smaller than the 'full' download you already have.

Another problem is the fact that there really are very few shades of gray in the images you are using although not as few as you are getting. Here are calibrated PNG versions of these files:

http://www.mmedia.is/bjj/utils/img2png/N1477488603_2.png
http://www.mmedia.is/bjj/utils/img2png/N1477490884_2.png

Also it might be a good idea to try this IMG file and this corresponding LBL file. This is a BL1 image of Saturn. It will be obvious whether it gets flatfielded or not if you sharpen it using an unsharp mask after calibrating and compare it to the uncalibrated version.

Let me know if this helps.

[tfisher]

There is a new IMG2PNG version available that should always complain when calibration files are missing:

Thanks. Actually, my major problem turns out I was opening the images as 8-bit instead of 16-bit png files. (Using ImageJ, which supports 16-bit per chanel, but requires a plugin to load 16-bit png files...).

Here's a sample of what I'm playing with; here is the Huygens landing site with a FFT bandpass filter applied and contrast adjustment: