(on vacation in Savannah, but secretly using the lobby computer...)

IIRC, for most of the images I played with, I did a Gaussian blur to help eliminate the "spottiness" of the color blobs. I didn't do any funky channel mixing but I did adjust the saturation and hue, then adding Ted Stryk's half-phase Mars image as the reference and adjusting the individual color channel curves to try to make it match Ted's image. If there was a cloud I tried to balance it out as a whiter shade.

Also, such an observation would probably best be carried out as a rapid-fire, single exposure setting sequence letting Mars slide underneath and capturing as many frames as possible. A dust storm or even high cloud cover would be perfect.


I was thinking two best exposed images and leaving the one or two underexposed ones out. What I imagine the team has problems with is using generic exposure settings for all observations as Mars' surface is backscattering and its brightness varies with phase angle. The several exposure settings are probably just safeguards against severe over/under-exposure

Bingo - we then sum the frames ( slightly under-exposed ones ) any features cancel themselves out- and we have a flatfield.

All really nice work! Where to start?!

Colour enhancement
Really cool enhancements of the colour of the images, we've been thinking for a while of boosting the colour to get more like the Mars people are used to seeing. If you get something you're happy with showing off it'd be great to post it to the blog and hopefully introduce it into our workflow so all images would be enhanced before upload.

Lens Artefacts
I'd love not to have the artefacts on the image and in fact I don't think we had them at launch (check out the pre-separation Beagle images). My theory is that debris from the ejection of Beagle (VMC sits right underneath the Beagle separation mechanism) might have hit the lens but I'm not sure it would produce the artefacts seen - anyone that can "reverse engineer" what sort of thing would cause them might make for an interesting blog article.

Flat-fielding
First off, I think the initial flat-field test by ugordan already seems to be a huge improvement on what we've got at the moment! I'm not sure if there are any images in the library that would at least get us closer to the data we need, perhaps some of the early exposure tests and crescent images or previous low altitude observations. Generally though we could consider doing a flat-field observation, although the chances to do low-altitude images are very rare so would be good to use them for proper imaging. What sort of image would we be looking for? Good exposure, under/over-exposure? Planet filling the image, only part of the image? etc. etc.

Exposure Settings
The exposure settings we use at the moment (the loop of 4) is as you guessed meant to be a good approximation to catch any lighting conditions regardless of phase angle. The middle two exposures are usually the best but I put the darker one in for extremely bright conditions and the brighter one for dim features like faint clouds. There have actually been two loops of 4 used: 0.4, 6.8, 14, 22.8 and the tightened range (currently in use): 2.8, 6, 10, 14 (all times in milliseconds). We actually have available a range from 0.4 to 191.6 milliseconds in steps of 0.8 milliseconds. If anyone can demonstrate from past images that tighter settings or even a loop of 2 or 3 exposure settings would still give good results then please let us know and we can change it! We have the ability to set the exposure individually for each image.

Image frequency
We've actually slightly improved the image frequency recently and are now really at the maximum frequency that VMC can support. This means an image every 44 seconds which is a hard constraint caused by the amount of time it takes the rather basic VMC to write the captured image to the on-board storage of the spacecraft.

Hopefully this answers some of the questions cropping up, of course please do send us by e-mail any results, processing or analysis done on the image for us to post on the blog. djellison sent us those wonderful movies he created and hopefully they'll be going up soon! Thanks again for your interest in VMC!

I think others here would probably agree that generating a useful flat field would be so valuable that it would be worth the cost of sacrificing "proper imaging" on a rare low-altitude imaging opportunity. The flat field would improve the appearance of every image in the catalog.

Here's a preliminary version of the VMC2RGB with flatfield support:
Click to view attachment
You'll need to place this file into the same directory as VMC2RGB.EXE. It's only a preliminary flatfield, has some artifacts and is cleaner in the lower center half of the image due to the pointing used in the images I used. I'm not really sure it does much other than reduce an artifact here or there and reduce color noise somewhat so your mileage may vary. I've seen older images that have one or two pixel offsets (probably readout errors or something) that completely negate the flatfield so be warned.

Wow, Gordan. Although I'm sure the flatfield can and will be improved, the difference that this preliminary version makes in the processed images is huge. Here's my best version before flatfielding (the first one) compared to a new version from after flatfielding. Artifacts remain, but they no longer distract from the actual albedo features visible in the image. Well done!
Click to view attachment
Click to view attachment

Holy **** - Awesome work. I'm going to have to do all my movies again when I get back from a few days out the office

I just noticed a few images here and there still have a 2-pixel shift to the left compared to the majority and thus the flatfield increases the noise. Take a look at images #10 and #11 from Nov 20 for example. I don't know if I'll be able to automatically deduce this from the image - seems like checking the rightmost pixel column might work as it appears zeroed out in that case.

In the meantime, you have been warned...

You might want to do a little filtering now. This is a GIMP unsharp mask, radius 40, amount 0.7.

Note that this is the only working wide-field camera at Mars until MRO gets fixed.

Fascinating to watch you guys working out the processing of these images.

There seems to be some subtle pale banding centered on the pole, more noticeable on the left half of the image. Am I correct to assume that these are real (high-latitude clouds), and not artifacts?

Cool, I'll try that out next, thanks for the suggestion.

Given its status as the only wide-field camera at Mars, is there something Mars Express could be doing with this camera that would actually have some value for weather monitoring, despite its limitations?

--Emily

I don't know enough about the particulars of when VMC can be operated relative to the eccentric orbit of MEx to know how close to global coverage it could get. It'd be interesting to see these images map-projected. I expect dust storms could be tracked, at least large ones.

Yet more great work, UMSF never fails to impress me! I've taken a look at possible opportunities and we may have some low altitude VMC opportunities but these are still not at our pericentre. At these altitudes we'd nearly fill the frame with the planet but there'd still be dark areas around the edges. Would this be sufficient to generate the improved flat-field? Also what are we looking to do? Blur the planet but keep the artefacts visible? I agree completely though that it's worth doing a flat field which will improve all future VMC observations, I was just playing devil's advocate Once again on that front, do we think a real flat-field observation would bring significant improvements on the already stunning improvements just shown by ugordan?

For the 2 pixel shift in some images I'll look into what might be causing that, could be a bug in the processing software I've written. Certainly in older observations there was an anomaly within the camera that caused sometimes severe shifts in the image pixels. This should have been solved for some time now though.

Finally, I hadn't thought about it but I guess we are the only wide-field camera currently in operation at Mars, although I read on the MARCI website that they'll start imaging again in December. While it'd be great to do more systematic weather monitoring we are quite constrained by the Mars Express orbit and primary science activities. As such that's why we slot into the apocentre observation slots where the spacecraft has just performed orbit maintenance. This gives us a gradually shifting phase angle on the planet (it's a crescent as you know at the moment) as Mars moves round the Sun. The orbit is roughly polar, highly elliptical and with a precessing argument of periapsis (for those that know orbital elements), which gives us the varying central latitude of the VMC images.

For the flat-field script update by ugordan, can we start using this routinely to process VMC images? Would be quite straightforward to slot into the automated workflow after image downlink. Also would be great if we could post a small article about how flat-fielding works and how you derived it for VMC (also an image of the flat field to see how messy the camera really is!). Thanks in advance!

Hope this helps with the info, let me know if you have more questions!

Since the flat field may improve with time, my inclination is that it should not be applied to the raw images posted on the website, though perhaps it wouldn't hurt to apply it to the PNGified versions? What do others think?

Yep, that was certainly my thinking, I want to leave the zip of the raw files exactly as-is, all the processing I do there is literally stripping the data out of the downlink packets, removing packet headers and splitting them into individual images. For the "instant gratification" PNG images though this could be a really nice addition. I also might look into systematically boosting the colour levels or sharpening as described if it's possible.

Just as some background...what currently happens is my Java software does the stripping described above (into the raws) then I call from Java ugordan's vmc2rgb v2 convert them into PNGs and then the Java packages it all up and posts it on the blog. The aim is to keep it "hands off" so no operator needs to be present and the time from downlink to being online can be kept as short as possible. Therefore we would need to work out a way to either implement any enhancement in my Java code or in vmc2rgb or a similar external exe that could be called from the command line.

Any ideas or suggestions let me know!

If it isn't a huge ordeal from your end to create one, (or maybe three) I'd say yes and I'm sure Gordan would agree.

The .raw file can be opened in Photoshop as a 640x480, interleaved 3-channel 16 bit image in IBM PC byte order. Here's a jpeg version:
Click to view attachment
It's actually the flatfield multiplied by 32768 (as pixel normalization factors are centered around 1.0) so as to fit into a 16 bit image and so I don't need to use 32 bit floating point images.


The way I created it was stacking a whole bunch (134) of recent images which weren't overexposed in Registax. This provided a good flatfield portion for the center and lower part of the detector because sharp limb contours infested outer parts of the stacked image:
Click to view attachment

I then applied a median filter to that image to remove most of the small scale details that affect VMCs noise the most. I then divided the two images and this resulted in my base flatfield - essentially only leaving the high frequency noise and manually taking care of the larger blemishes present. I used another older observation that had Mars fill most of the screen at low phase to produce another flat in a similar way and to mask out the circular arcs in the lower part of the image edges and in the upper portion. It's not a perfect result and rings are still plainly visble in the flatfield above, but I don't have a good way of getting rid of those. Making flatfields is not really my area of expertise.

The caveat of me using a median filter is that only the high frequency noise is flatfielded out, so the flatfield does not account for any vignetting and/or large scale mottling. I don't know how to work that out from the data at hand, hence the flatfield above looks uniform brightness-wise.

If someone is willing to produce a better flatfield, I'd be more than happy to see what can be done.

This flatfield looks very promising Ugordan. Thanks!

I send one image from my "ping pong" with colors (inspirated by Elakdawalla).
This one is yet without any flatfielding

I think having the unprocessed raw images available is necessary, while having the pre-processed PNG images available will give instant gratification. (It will be necessary to show what version utility was done to process the PNG, though)

It might be handy to have a direct link on the Mars Webcam website to the latest version of Gordan's vmc2rgb utility as well as a general article on how flatfielding is done And including a link to the lastest flatfield image version will also be neat.

Here's an animation of some data from quite a while ago (September 2008) which is greatly improved by the flatfield. The frames are registered on the keyhole-shaped region below and to the right of the large crater in order to better show spacecraft motion.
Click to enlarge (1126x830, 2MB):

Very well done, and an awesome animation!

How is that application going Doug? The image work appearing here now is astonishing and the pro/am collaboration definitely newsworthy.

Digging through the archive (http://webservices.esa.int/blog/album/6/57):

Might be useful for making a dark field:
Sept 19, 2007 images: (http://webservices.esa.int/blog/album/6/298)
May 25, 2007 images: (http://webservices.esa.int/blog/album/6/121)
January 28, 2007 images: (http://webservices.esa.int/blog/album/6/108)


Might be useful for making a flat field (close-up images where features drift across field of view, more or less filling most of the view):

July 1, 2008 images: (http://webservices.esa.int/blog/album/6/217)
July 5, 2008 images: (http://webservices.esa.int/blog/album/6/222)
July 12, 2008 images: (http://webservices.esa.int/blog/album/6/225)
July 16, 2008 images: (http://webservices.esa.int/blog/album/6/227)

and June 22, 2008 (sequence animated by Fran): (http://webservices.esa.int/blog/album/6/214)

This was in fact the June 22, 2008 data referenced by Juramike, and September comes from me lazily using the "date modified" of the images instead of looking up the actual DoY.

Here is yet another "super-resolution" attempt with the MEX-VMC image sequence taken on July 5, 2008 of Valles Marineris:

Full image corresponding to Image No. 2:
Click to view attachment

Subsection of Valles Marineris:
Click to view attachment

Even closer zoom to E edge of Candor Mensa (image scale is 1 pixel per km):
Click to view attachment

To make the super-res, each VMC image was bicubic enlarged 10x, then unsharp masked, then dropped down 50% (thus 5x). A really complex series of masks was used to allow the useful parts of the overexposed images to overlay with a darker "base" image. The images needed to be grouped carefully, since changing perspective made the overlay difficult.
(So...image 1 and 3 were overlapped with 2, while image 7 and 4 were overlapped with 6.)

It was a heckuva lot of work, and while a very slight sharpening can be seen in at the pixel level, it wasn't worth the effort.

(The final super-res image has a pixel scale of 1 km/pixel).

Color study of Valles Marineris image:
Click to view attachment

Had a go at a few more since the flatfieldorama version of vmc2rgb (only one request - can we have a version that would work with a batch file that reads vmc2rgb.exe *.raw for example, like img2png? )

I'll echo that request unless someone can tell me a quick and dirty way to run an EXE on all items within a folder...

Handel wrote the Hallelujah Chorus for this exact moment.

LOVE the movies.

I made an LRGB image to try to enhance contrast in the Valles Marineris image.

To make the luminance layer, I used the "Emily" channel mix and looked at a bright region and a dark region to try to find the optimum combination for the black and white conversion. A bright region "box" was defined around Candor Mensa and the median and standard deviations of the pixel values selected. A dark region box was also selected, as well as a good variance box.

Click to view attachment

Examining the pixel values, the standard deviation inside the bright or dark boxes is about the same in all the channels (good!)
The standard deviation is the highest in the Variation box in the red channel.

The red channel has also the largest difference between bright and dark pixel values, followed by the green channel.

From all this, it looks like the most information is in the red channel, followed closely by the green, and the blue channel is kinda worthless in the chasms.

So the black and white conversion weighs heavily on the yellow and green, followed by the red. (Ideally, more red should be used, but fiddling revealed that the center regions would blow out quickly).

So the Black and white conversion layer settings to create the luminance layer are:
Red +60%
Yellow +70%
Green +70%
Cyan +20%
Blue +50%
Magenta +50% [not much in the image at all]

-Mike

Here is the RGB image compared with the HiPass Superresolution LRGB image in the area near Candor Mensa.
(HiPass 30 pixel, color layer gaussian blurred 5 pixels)

Click to view attachment

The color is less "speckly" in the LRGB image, probably due to the Gaussian blur of the color image.

-Mike

Using the filenames, it seems images less than 116 seconds apart can be reasonably overlaid.

Click to view attachment

For images greater than 232 seconds, the spacecraft motion (at this altitude) causes enough of a perspective change that manual warping is necessary to overlay similar images. For example, images No.2 and No.6 were significantly different.


Here's an attempt at a super-resolution combination using only Image No. 2 and a manually warped Image No. 6:
Click to view attachment
The subtle differences even due to my limited warping capability are enough to make the lower dark patch blurry.

Bottom line: group files that are within a max range of 120 seconds.


[On a really close pass, it might be interesting to have a rapid fire cycle of 3 or 4 exposures of the same duration, then a second set of 3 or 4 at a different exposures of longer/shorter duration. Then a super-resolution image might could get made.]

Final mosaic of Valles Marineris prepared from MEX VMC images:

Click to view attachment

Much higher resolution (2 km/pixel) TIFF image here [5.5 Mb at highest resolution]: http://www.flickr.com/photos/31678681@N07/4152369836/

-Mike

Nice work! All this is making me miss the days when I had time to process images

Here's a blink comparison between a Viking 1&2 mosaic (PIA00422) and the MEX-VMC HDR Super-res HiPassLRGB composite. Features as small as 20 km can be discerned in the MEX-VMC composite.

Click to view attachment
(Animated GIF - click to animate)

Static view:
Click to view attachment

-Mike

[EDIT: Added an annotated version of MEX VMC composite with 26 features added as notes on flickr page: http://www.flickr.com/photos/31678681@N07/4153691165/]

My lateness to this particular party does not make my agreement with this statement any less hearty.

Though I think right now it's that massive C major chord from Handel's "Creation" blasting in my head!! and there was....LIGHT!!!

And...there....were..... FIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIILES!

(I always resented the bass-part going down on that cadence rather than up)

I've gone silent on this topic because I'm in the middle of a big project with it -- I methodically downloaded all the RAWs, then used Helvick's trick and Gordan's executable to convert them all and now I'm inspecting every set and taking notes on which ones are worth processing. I'm about halfway through that process. Ask me very nicely once I'm done and I may zip up the "worth processing" ones into a single downloadable file

Hi all!

Still very much looking forward to seeing the results of the processing efforts going on but just wanted to give you a heads up on some images dumped this morning. We got an opportunity to do a low-altitude limb pass so caught quite a few shots of the planet limb and there appear to be some very interesting clouds on these images. You can take a look here:

http://webservices.esa.int/blog/post/6/932

Enjoy!

Nice!
Three sets in one day (14.12.2009), I think, this is new record!

Looks like Emily finished her project. Awesome poster!

Thanks! For some reason I didn't notice until just now that I'd managed to post a version with the space not yet adjusted to be black. I've fixed the version posted in the blog, am working on the Cafepress version now.

Wow, 64 images! amazing!

Repetitive work is my friend I spend what little kid-less time I have thinking about how to develop the workflow, then while kids are around and are almost entirely distracting me I can perform a repetitive task with almost no brain cells focused on the problem.

Good, because repetitive work is my enemy. I would have gotten bogged down in the idiosyncrasies of individual images and never finished.

09_352 (Dec 18th) Low Flyover Mosaic - GREAT sequencing boys, I think there's possibly a better flatfield to be made out of this as well!

09_349 as well...