I've tried to do just that, but as of yet such attempts simply break my poor little pc ;-). Variance works okay but my FFT code needs some improvements to make it work with such a large dataset.

I believe the trick of first calculating average brightness and then relating each pixel to this average is certainly worthy of following a bit further to see where it leads us to, but take in mind that it has a disadvantage too: by working with 'relative' data you can only compare data (for drive-analysis) within the same picture, you can no longer state that 'green is maximum drive distances', you can only say that it is 'best possible drive distances in this picture'. But maybe that's what it has been saying all along...



Note that it looks like this region is also where there seems to be a significant difference in HiRISE and CRISM data (see my previous message). Maybe it points to something, maybe it doesn't...

What I'm trying to do at the moment is see if I can create a large, grid-bound, dataset (multi dimensional array) of the victoria-endeavour area incorporating not only HiRISE but also data in other spectral bands. Basically any data should fit as long as its chart-aligned (fits to a lat/lon grid). In the most ideal situation it should even be possible to incorporate navcam/pancam data (convert to polar view, then given a known rover-position add a lat/lon grid). Lots of work but basically it can be done. Once I have this dataset (and probably it will be several datasets for the whole area is much too large) then it should be possible to combine, compare, and analyse data from several different sensors/spectral bands. See where this takes us. Don't expect this to be finished tomorow however ;-).

Regards,

Geert.

ok, I think the "texture classifier" approach should be worth a try
Here is an attempt to create a map (of the area south of Victoria) with a multi-scale bank of texture filters (my own variant of "mini-Gabor" filters, basically designed to capture granularity, directionality, edgeness and other micro-texture features).



There are 13 filters applied to each 9x9 pixel moving window ( window-spacing = 2 pixels, with the rest interpolated by bilateral upsampling).
To capture features at multiple scales I decompose the original image in a gaussian pyramid and applay each filter to each pixel at each level of the pyramid,
So altogether we have 13 filters at 7 spatial resolution scales.
I used a PCA-based mapper to reduce this 91-dimensional original feature space into a 2D color space (LAB) that can be used for visually analyzing the resulting texture-map overlayed with the brightness information of the original bw image.

As this is only intended as a first test, I did not try to assign the color-mapping to any meaningful scale in the sense of "dangerous/easy".
The different color just represent the wo most significant dimensions in over-all textural variation as it results from the PCA 91-to-2 dimension reduction.
(coincidentally, a quick Eigenvalue-Analysis shows that two dimensions are already sufficient to capture about 80% of all the variation...)

RESULT:
At a first glance, the usual "terrain classes" seem to be distinguished: "red color tones" = "soft sand", "blue/cyan=nort-south-trending ripples", "orange=east-west oriented ripples", "greenish = bedrock ?", and so on ...
Now, of course, it would be the task of the geologists and rover driver specialists to assign the colors to "meaningful" terrain classes.

For this first test I applied the algorithm only to a reduced 2000x1000 pixel crop of PSP_009141_1780 which took about 200 Seconds running time.
So in principle the approach should be feasible to be applied to full or almost-full resolution imagery as well (thanks to the implementation in good old plain C


If time permits I'm going to download the whole HiRISE-JP2 tonight and try another run at a finer resolution level ....

P.S.: I have no Idea if this is useful at all, just an attempt to see what the incorporation of multiscale texture algorithms could add to the existing brightness based
analysis ...

Wow, this is a pretty good result! I will have to grok this.

Paolo

Here is an new version of the map where I omitted the four highest levels of the gaussian pyramid thus putting a little bit more emphasis at the small scale features:




( Unfortunately I don't have nearly as much time for all this fascinating stuff as I wish I had

And yet another version of texture map. This time I used the full three dimensional color space to visualize not only two but three dimensions of the original PCA-projected multi-dimensional feature space (which, however leaves no dimension for the brightness channel anymore

Also, I used a 12x12 pixel moving window and restricted the analysis to the finest resolution pyramid level.

Absolutely fantastic!

From a geology point of view, I think it is really neat (and significant) how you've been able to light up the EW (WSW to ENE) trending ripples.
Those are more likely to be the most recently emplaced stuff, and could be a potential problem if it is indicating looser dust.

(And I don't think Oppy has traversed any of these type of areas yet, the closest she got was down near the floor of Endurance)

Is there any way to classify and convert this to a grayscale? It'd be neat to try to see how it correlates with past and future observables.

-Mike

There seems to be a similarity between this analysis and the CRISM IR image.

Both show a much more distinguishable area in the southwest (dark red in CRISM and light purple in yours) and a less accented area in the south and southeast (the green 'passage' in CRISM corresponds with the black area in your analysis). This seems less obvious in other analyze types, hard to say what causes this.

Regards,

Geert

Slowly getting somewhere in trying to combine HiRISE terrain calculations with data from other instruments.



Above an overlay of the area immediately south of Victoria with relative infra red brightness as measured by CRISM.



Same area, however now with a color overlay of Mafic mineralogy (ir_maf) once again as measured by CRISM.

All as yet very preliminary so tread with caution, due to the various scales and map-projections of the data it is a constant wrestle to convert all data to one and the same chart grid and data format. Positive thing is that once I've got it into the dataset and the correct grid, all further image calculations are very fast and easy.

Regards,

Geert.

Excellent work, Geert ! (I know how time consuming this image processing can be ....)
A step forward in combining as much individual spectral/textural "bands" into one analysis as possible ..

In the same spirit I am now planning to incorprate the Fourier-based Analysis into my texture-analyzer.

Question @ James Canvin:

As I have not followed the details of the former discussion: could you provide the details/parameters of your latest Fourier mapping algorithm ?
(local window size, which components of the FFT used (amplitude, phase) ? pre-processing used, other parameters ?)

Thanks
Bernhard

I'll have to go through my code later to remind myself.

But it is basically the maximum of the Fourier Power Spectrum on ripple scale lengths (~8 - 28 pixels I think) for 1D East-West slices 32 pixels long, each value being an average of ~4 pixels North-South. iirc.

In the search whether chemical surface composition has influence on 'drivability' of terrain, below are compositions from CRISM data and the wellknown HiRISE images of the area around Victoria crater.



ferric minerals



low-Ca pyroxene



high-Ca pyroxene



variety of iron minerals




olivine or iron phyllosilicates

All of above is in 10 mtr resolution (and that's already a extrapolation from CRISM), no use to try to narrow down the area any further.

If we compare this to the earlier study's of terrain it looks to me like there is clearly some correlation between chemical surface composition and general terrain as measured earlier. When I find the time I see if I can create a similar dataset for the remaining areas covered by MER and let the computer crunch a bit more on correlation factors and such. It's a lot of work but results are interesting.

Regards,

Geert.

This is a fascinating line of inquiry. If you're right then route choices made purely on drivability criteria would result in a degree of observational selection in the suite of minerals encountered by the rover. Of course it could also work the other way, with the routemasters deliberately seeking out chemically distinct areas from the CRISM maps to get the most representative picture. I wonder which it is?

Maybe, but what I'm trying to do goes back to a bit earlier in this thread, when we had a discussion whether sanddunes were associated with certain chemical compositions. If you look at the terrain, there are parts where there are (relatively) big sanddunes, and other area's where there aren't. I can imagine that one of the factors might be the composition of the sand itself, whether or not it easily 'moves'. This is a hypothesis which can be checked by comparing maps of chemical composition with maps of dune-locations, as I have just done.

An other item might be the driving itself, how far do the wheels 'sink' in, how much grip do they have, etc, etc, this might also be related to the chemical composition, with Mike's dataset of driving conditions it might be possible to see if there is a relation there also.

And finally, it has been worrying me a bit that during this whole thread on drivability analysis we have so far only been looking at the HiRISE images, and then mainly at the brightness (no matter if you analyze on variance or FFT or any other combination, the original input is always pixel brightness), so I like to check whether there are other instruments/datasets which can be used to check the results of all the analyse-techniques (Mike has already done a wonderful job in relating the outcome to the actual driving conditions, which allows to check the validity of any hypothesis).

Regards,

Geert.

THEMIS IR-night would be a good start - Sure, it's only 100m/pixel, but we're talking about 15km here.

Correct, THEMIS IR I have also already in the correct format in the dataset, however the "low" res makes it harder to check the correlation with dune size.



Above is the Victoria area at 10 mtr res with an overlay of THEMIS Night IR combined with CRISM IR, however this is a preliminary version, I can probably improve a bit further on it by playing a bit more with the settings.

I have also TES data now in the dataset, and even some Viking data although that's even less resolution. I'm still working on the ESA data, there is data but the dataformat is a bit harder to fathom...

Anyway, I think there is quite a lot of material to see whether we can find a correlation with our own previous terrain-models, lots of data to crunch ;-).

Regards,

Geert.