Very good job Bill!
QUOTE (wbutler @ Sep 30 2008, 06:44 AM)
wstdev - the standard deviation of the amplitude distribution, which should be related to the width of the distribution - bigger values should mean more frequencies and thus more variation in ripple size within the window (no reciprocal here)
That is indeed exactly what I'm doing with my own tool, basically it should tell you something about the roughness of the terrain, a lot depends however on how big you set the window space and the sample rate within this window. It sure seems to mark out the bigger ripples and the more rough terrain but it is difficult to note which terrain will give most problems to oppy.
QUOTE (wbutler @ Sep 30 2008, 06:44 AM)
wangstddev - the standard deviation of the amplitude weighted angle distribution - bigger values should mean more angles in the window, which was my attempt to find the curvy ripples. This didn't seem to be successful.
See my earlier attempts at finding ripple angles via the same method, If you take a very big window (a big distribution of ripples) it averages out nicely on the average ripple-direction, however it is very hard to get correct directions for each and every ripple as this method very easily gets confused and to pick out each and every ripple correctly you need to set the window quite small and that doesn't give you a big enough distribution for the tool to work on, at least that seems to happen to me.
What I
hope will give you an indication of 'cross-ripples' is taking the standard deviation in the frequency range of the ripples, so instead of the amplitude distribution, you try to find an average wavelength for the ripples across the window, then take the standard deviation from the frequency spectrum, a wide spectrum would mean a great variance in different wavelength, and so a bigger chance of cross-ripples and chaotic terrain... As yet I have not been able to do this as it takes a lot of computer-power to do this accurately.
QUOTE (wbutler @ Sep 30 2008, 06:44 AM)
I have also tried some classifications, with mixed results. The small ripples (eg in the center of erebus) are easily distinguishable, and the medium sized ripples on the east side of the crater are reasonably distinguishable. But I have not been able to highlight the curvy ripples, or reliably segment out the bedrock areas yet.
Correct, I run against exactly the same problem. Bedrock I can find if I set a small window (small distribution) as then its variation doesn't match the surrounding ripple-fields and it will get classified as 'flat', however if you set an average 64X64 window the bedrock areas are just too small to get noted. The dangerous curvy-ripples are still alluding me...
In an other post it was mentioned to try some 'fingerprint' trick, in other words pick out a sample of a 'dangerous' ripple, then compare this across the image to spot the dangerous ones. I'm going to try to do this, but I've never written FBI-software so I don't know how long it will take to get that trick working ;-). As is often the case, the basic idea is quite simple, however the technical implementation is a different story, you'll have to take care of image-scale, different viewing-angles, ripple-directions, brightness, etc, etc, and then compute more of less a figure which tells you for which percentage the sample corresponds to the 'dangerous' ripple... It sure looks nice on TV, finding a face or a fingerprint like that, and I'm sure it
can be done, but my time is limited and this is something I've never tried before, still it's fun to do...
QUOTE (wbutler @ Sep 30 2008, 06:44 AM)
I think that Juramike's idea of looking to the ground truth to choose classification examples is the right way to go, but of course Paolo is right that that job is best done by the guys with the most data. I would be happy to incorporate such guidance in the classifier.
I think I've done about all I can at this point without knowing more about what is helpful and what is not. But it has been a very interesting exercise. I was not optimistic at all in the beginning about the 2dffts, but they turned out to do a very nice job of distinguishing ripple spacing! Looking forward to the trip!
I also think at a certain moment you more or less reach the limits of what you can get out of these HiRISE images by computer-power, it's fun to try the 'fingerprint' thing but by the time I've worked that out probably anyone who looks at the pictures which his/her own eyes has already spotted all the 'dangerous' ones. Exactly the same can be said about 'dust-traps' and such, without multi-spectral images I have not the faintest idea how to spot these using software although they are 'obvious' to anyone looking at a picture...
As has been said often before, in the end it's the human eyes, and the experience of the drivers, looking at the pictures which are required to find a route, it's a lot of fun to see how far you get with software-processing but with only the HiRISE images there's just no way you can ever get anywhere close to the true experience of the drivers and their eyes.
I think we more or less reached the end of software-processing, and thanks to all the tremendous work in comparing results with ground-pictures we now know more or less what each trick will tell us. As soon as the 'new' HiRISE images arrive we can loosen our tools on them, but most of all we will need people with experienced eyes, just looking at the pictures and simply marking out the 'dangerous' spots.