Endeavour Drive - Drivability analysis Options

[Fran Ontanaya]

Grain sorting may be a factor. The albedo of the dunes won't be constant if they are topped by dust. Maybe that's why bigger dunes are brighter on average.

[fredk]

Following up on my previous set of plots, here I've turned my final variance plot above into a colourmap for the original greyscale image. Green means go, red means don't go, and purple is bedrock:


The red strip along the bottom and right edges is an edge effect of the processing - it can be cropped away. Again, I have no idea how robust this would be - it would need to be tested on a much larger area. And now I really must get back to work...

[jekbradbury]

Now, who wants to do the same with these images?

http://www.unmannedspaceflight.com/index.p...y&pid=76383

Here is the result of a variance-based algorithm on the most ripply of those images:

[ElkGroveDan]

I took a shot at Fred's image and this is what I came up with

Attached thumbnail(s)

 

[Tesheiner]

Could you guys tell me what are you smoking?

[TheChemist]

Doug is gone for two days and this place turned into an art gallery

[Shaka]

Don't anyone let Dan cut off his ear!

[john_s]

Here is the result of a variance-based algorithm on the most ripply of those images:

That's starting to look very interesting! Can't wait to see the other images treated this way. I would only suggest (now we're in an artistic mood) a more intuitive color scheme- this one has red at both ends of the scale. Maybe green colors for small ripples ("Go!") and red for big ripples ("Stop!"). And of course the scheme should be consistent for all the sub-images.

John.

[stevesliva]

Did it have red at both ends?? No wonder I thought it wasn't handling bare rock well...

[RoverDriver]

Following up on my previous set of plots, here I've turned my final variance plot above into a colourmap for the original greyscale image. Green means go, red means don't go, and purple is bedrock:


The red strip along the bottom and right edges is an edge effect of the processing - it can be cropped away. Again, I have no idea how robust this would be - it would need to be tested on a much larger area. And now I really must get back to work...

These maps are pretty good. I did not expect to work this well since the variance filter is isotropic while the ripples clearly have a preferred direction. I would have expected some confusion between ripples and outcrop. Pretty amazing.

Paolo

[Juramike]

Here is the result of a variance-based algorithm on the most ripply of those images:

Art AND Science!

I like it!

Since the ripples have a preferred orientation, would it be possible to do some sort of transform in a direction 90 degrees to the original? Then stack the images? Sort of like a polarizing filter. Objects that don't have the linear variation orientation of the dunes would end up getting lit up - e.g. bedrock?

(Then you could do the dunes in the cold-colors to warm colors scheme, and indicate bedrock as shades of gray.)

-Mike

[jamescanvin]

Great work with the variance method, amazing.

I've just about got my Fourier method working. Here is a strip of the HiRISE image I've been testing on, chosen as it has some serious ripples on the left and very small ones on the right.



Looks pretty good so far

James

[ElkGroveDan]

Kudos and high-fives to everyone (but me) for their ingenuousness. However, I for one can't wait to see James' final product.

[jekbradbury]

The "red at both ends" issue was an artifact of post-processing. Here is a fixed image:



If anyone could point me in the right direction towards being able to use the JP2s with ImageMagick, that would be very helpful.

[Juramike]

I came up with a low-tech method that seems to work:

1) Take image (I used fredk's image as a test); duplicate layer:


2) Shift one image 3 pixels to left (crests of big dunes are 7 pixels apart)
3) Make difference of the two images (this will put crests in troughs and make overall brighter):


4) Gaussian blur 5 pixels:


5) Massively increase the contrast (using levels):


6) Merge layers then convert to colorized image (dark = good = green; bright = bad = red):



7) Overlay colorized image with original Hi-Rise image (the 5 pixel borders on left and right are lost):


The beauty of this method is that even bedrock areas that have big dunes in them get indicated as bad.

-Mike