For those who have GIMP, what about using "Selective Gaussian Blur"? That only blurs low-contrast areas... I'm testing that out as I type this.

Erebus-to-etched terrain shifted-differenced-multiply blurred-contrast enhanced-colorized-and overlay/multiplied image:


10 pixel shift to left (in this image, which I think is higher res, the biggest dune crests are 20 pixels apart).
Multiple gaussian blur (I forgot how many)
Contrast level setting is pretty subjective I cut to about 80% of histogram as pure white, 20% of histogram as pure black
Coloring is also pretty subjective (I put yellow as the mid-point)
Overlay using 100% opacity and multiplication of layers.


-Mike

It's funny how everyone's using my IAS Viewer screenshot as their test image. That was essentially chosen by clicking randomly with the zoom tool until I was at 100% scale. I prefer using the pre-Victoria terrain, since we know what it looks like.

I went ahead and implemented a variation on Juramike's first idea:

Ha! We posted at the same time!
(I think my version is more pessimistic!)

-Mike

In my version, green doesn't necessarily mean go. Only blue (or, really, purple) means go. Green means driveable. The nice thing about it, though, is that someone with more knowledge of the rover's route through the pre-victoria dunes could cross-reference the colors to the ground truth and get a more accurate answer.

Simple brightness map.



I did:
Duplicate background - Pixelate 20px - Transparency 33%
Duplicate background - Pixelate 80px - Transparency 66%
Duplicate background - Pixelate 160px - Transparency 100%
Combine
Normalize
Gaussian blur - 80 px
Color map from gradient (Tropical Colors)
Normalize background
Move the background on top and change mode to Combine grain

Could someone do the same (break up the huge JP2 file into manageable pieces) for the recent HIRES image of the terrain south of Victoria.

All these colorful HiRISE maps are cool, but more work needs to be done on the algorithms. Some of the places marked in red - for stay away - are actually the best places to go. It's hard to see some of that using my XP machine; but when I check it out on my Vista computer, it is very clear. You can see things like 10 times clearer on Vista using the same program on a comparable computer.

From the looks of it, this is going to be a difficult algorithm to write and make accurate. Good luck.

I'm tempted to do it by hand myself, but don't have enough time.

Thanks for the clue, Shaka. I found myself scratching my head while looking at Dan's work, wondering how that scoundrel achieved such a result. It's all so obvious, now.

To the rest of you producing those amazing terrain recognition maps, keep it up. I'm seriously impressed by your ingenuity and I'm really excited with the maps I am seeing so quickly. This is one of the most interesting threads we've had around here in quite a while. I've noticed that some of the methods seem to do a better job of flagging the dangerous half-pipe pinch-offs as red territory. Were not Purgatory and Jammerbugt such geometry's? This exercise is not simply about large versus small drifts.

I'd like to echo previous comments that we should adopt a standard color scale for these contour maps. That would make comparisons easier. Its probably a bit early to make judgments regarding the different methods being used, but so far, Fran's first color map is my favorite. The contours really seemed to hug the terrain changes.

Test pics on GIMP - can someone tell me if these are good or not?

What I did was start out with a very light Gaussian blur, then use 'selective Gaussian blur', which only blurs the low- or no-contrast areas. I used this on the picture until the flattest terrain in it was monochrome gray. Then I used 'fuzzy select ' to select the entire flat region and colored it pale green (hue=125, saturation=100, lightness=70 in case anyone cares). I then took the unaffected (contrasting) region, used 'color balance' and pushed all three ranges (shadows, midtones, highlights) all the way (100) to red on the red/cyan axis.

Green = flat, dark red = crater bottoms, middle red = slopes, light pink = hilltops.

Problems I've noticed: crater bottoms, especially shadowy ones, look worse than they are. (I hope Oppy isn't driving through those anyway!) Also, the amount of 'selective Gaussian blur' is subjective. It's hard to select every tiny color dot in the more complex pictures - especially inside craters.

I performed this process on two of the old route pics suggested two pages back - "backshell to Endurance" and "Beagle to Victoria".

Please tell me if this is any good, or whether I should just leave it to the experts! If you like these, I'll do it on some pics for the route to Ithaca/Endeavour.[attachment=15763:backshel...ndurance.jpg]

I don't see any gross errors made by mine, but if anyone spots any bad points then let me know. I was using a very simple extraction of data from the Fourier power spectrum so I'm sure there is room for improvement. Determining what is a useful measure is not easy however, so pointers to problematic areas would be useful.

Hopefully I'll have a LOT more data for you to look over this evening (GMT) - maybe even an entire HiRISE image!

James

Manageable JP2's or another format?

I'm going to be splitting the JP2 into smaller JP2's for processing and could put them on line If required. I don't know much about converting them to other formats though - especially if you want to keep the full 16bit depth.

James

Bingo.

Duplicate layer
Gauss 5px (small ripples are washed, dune shadows over exposed bedrock are brightened)
Difference mode
Combine (results a very dark layer)
Raise both brightness and contrast to 120.
Gauss a lot
Apply gradient map red-yellow-green.

I don't really remember all the sol-by-sol driving history as we went towards Victoria Crater. Is there a better HiRise image that would make a good test?

Something where we could overlay Oppy's path and use the map-to-field information to set colors that indicate doable and not doable based on Oppy's images (and driving experience). (Purgatory definitely needs to be "red")

We also need an example of "doable" bedrock and "bad half-pipe pinch-off" bedrock in the same image that could serve as a test.

-Mike

I don't think we need, or even want, a method that tries to classify every ripple and looks for dead ends, etc. What I am trying to produce is a map that says 'Over here the ripples are generally small - you could go around any rouge Purgatories without too much difficulty' and 'Over here are a load of monsters - stay away or you could spend weeks getting through the maze'

A Purgatory type ripple doesn't need to be 'red' if it's easy to go around.

After all, this map is to help get a rough idea of the route Oppy should take to get to Endeavour the fastest - not to find every little danger to avoid (That will be Paolo's job as we go along!)

James

P.S. That's great work Fran - my only worry is that the exposed bedrock is always 'green' I'll be interesting to see how this works over the larger variety of terrain seen on larger scales.