Just a test - 64K Moon

[cartrite]

Just to make sure we are talking about the same stripes, here is a 2k version that was not filtered.

cartrite

[cartrite]

I'm not sure what to do with the null areas yet. And the 64k file may require a wider lowpass boxcar. It should be a little wider than the actual stripes. The lines should probably be the same amount of the samples for the lowpass program (square boxcar for lowpass). The highpass will always have 1 sample and the same number of lines as the lowpass.
cartrite

[t00fri]

Yes, Steve,

that's the stripes. And yes, the boxcars have to be rescaled with the resolution eventually. But if we manage to get rid of the artefacts for the 8k texture, we are certainly on the winning track...

Fridger

[cartrite]

For the 8k file, this about the best I can do with the filters.

qview.jpg

I used this code to run the ISIS programs.

Code: Select all

lowpass from=Clementine8k.cub to=Clementine8k.lowpass.cub samples=53 lines=53
highpass from=Clementine8k.cub to=Clementine8k.highpass.cub samples=1 lines=53
algebra from1=Clementine8k.lowpass.cub from2=Clementine8k.highpass.cub to=Clementine8k.filtered.cub operator=add

If I use smaller values, the stripes become visible.
cartrite

[t00fri]

Hey Steve,

congrats. That result looks pretty cool! I knew you would not give in before you hacked the problem

Excellent.

Now we simply have to rescale your boxcars for the 64k image!

Then come the black bars...

Fridger

[t00fri]

Steve,

after just running your filter, it seems however that quite a lot of detail is lost as well.

I'll try with 53 --> 37, which is the "theoretical" width.

Fridger

[cartrite]

Are there any other 64k data sets of the moon? Even a scaled up 32k could be used underneath the Clementine image. If the contrast and brightness between the 2 can be matched, the null areas would be filled in with automos. Just got to use the Clementine image on top.
Did you see the 4k image?
viewtopic.php?p=115375#p115375
That used 151 samples/lines for lowpass.
Seems like a lot of smaller stripes on a 45 degree angle crossing each other.
Just seen your last post. I'll try a rectangle boxcar with more lines than samples. Something like 53 samples 501 lines for lowpass, and 1 sample 501 lines for highpass. Did you try the program cubenorm?
Nevermind, cubenorm didn't do a thing.
cartrite

[t00fri]

Are there any other 64k data sets of the moon? Even a scaled up 32k could be used underneath the Clementine image. If the contrast and brightness between the 2 can be matched, the null areas would be filled in with automos. Just got to use the Clementine image on top.
Did you see the 4k image?
viewtopic.php?p=115375#p115375
That used 151 samples/lines for lowpass.
Seems like a lot of smaller stripes on a 45 degree angle crossing each other.
Just seen your last post. I'll try a rectangle boxcar with more lines than samples. Something like 53 samples 501 lines for lowpass, and 1 sample 501 lines for highpass. Did you try the program cubenorm?
Nevermind, cubenorm didn't do a thing.
cartrite

I am not sure about other 64k Moon data, since I had a pretty long break between doing some work with the Moon . In general it should be rather straightforward to replace the NULL values by the average of the surrounding boxcar. BUT for bigger black bars the average will also be black .

After you previous mail I run the lowpass with (37,37) and the highpass with (1, 501) which seemed to look more detailed....

No I did not play with cubenorm. But I had read it wasn't effective

Fridger

[cartrite]

I don't think ISIS treats null values as valid numbers in equations. They may be ignored. In automos, null values are transparent and the mosaic that is beneath will show though.
I just tried lowpass 63 samples 201 lines and highpass 1 sample 201 lines. That looked a bit better than the boxcars I used earlier.
I thought I read somewhere in ISIS documentation that vertical stripe corrections should share a common value in the line parameter to keep the resolution loss to a minimum. Horizontal corrections share a common sample value. That could be a bad memory though. My memory is about as bad as my computers.
cartrite

[cartrite]

I notice that the Clementine data is a lot darker than Celestia's moon textures. I did a brightness and contrast enhancement of 60 units in the Gimp and it was still darker than Celestia's moon.
cartrite

[t00fri]

I don't think ISIS treats null values as valid numbers in equations.
Near the bottom of the GUI there is often a replacement option of the NULL values.

I just tried lowpass 63 samples 201 lines and highpass 1 sample 201 lines. That looked a bit better than the boxcars I used earlier.

I just tried that combination and I think it REALLY looks like an improvement!
Let me try to scale this up to 64k and run your script Just a few minutes to go...

I thought I read somewhere in ISIS documentation that vertical stripe corrections should share a common value in the line parameter to keep the resolution loss to a minimum.

Yes I can confirm this. But I also remember that one dimension should completely contain the stripes in question, hence for 8k and vertical stripes, theoretically that dimension should be pretty big.


Fridger

[t00fri]

I notice that the Clementine data is a lot darker than Celestia's moon textures. I did a brightness and contrast enhancement of 60 units in the Gimp and it was still darker than Celestia's moon.
cartrite

Yes, I agree. It's pretty dark. I read in the ISIS forum that there is a normalization routine for this. But I forgot its precise name.

Fridger

[cartrite]

I think it is photomet but this was already used on this file. There might be some other program to brighten the image though. I also tried to change the albedo setting in solarsys.ssc but that didn't do much.
cartrite

[t00fri]

Trara!

Steve, with scaling your boxcar up by a factor of 8 relative to the 8k experimental Moon file, the stripes are gone on the BIG 64k Moon while sharpness/detail has been nicely preserved!

Here are two examples:

The total Moon without the previously well conspicuous stripes.On the lower left you see that infamous faint, lengthy black line, well familiar from all Clementine maps. We'll get rid of it soon...



Here you see in case of the Schroedinger crater that all detail has been beautifully prepared in the cleaning process!



+++++++++++++++++++++++++++++++++++++++++
PS: Yes with the recent script, it only takes 8 minutes and ONE click to do the full 64k VT set!

After a little more cleaning, I shall make the 64k binary input file (Clementine) available for download at CelestialMatters.
++++++++++++++++++++++++++++++++++++++++++
Enjoy,
Fridger

[cartrite]

Looks nice. I'm running the filters now. I'm using samples=253 lines=801 for lowpass and samples=1 lines=801 for highpass. Running algebra now.
Here is another thought. Run automos with the lowpass cub below the Clementine64k cub(algebra cub). Only the null areas will have the lowpass cub show through.
This will fill in the null areas.
Did you apply a special "stretch" when running isis2raw?
cartrite

[t00fri]

Looks nice. I'm running the filters now. I'm using samples=253 lines=801 for lowpass and samples=1 lines=801 for highpass. Running algebra now.
Here is another thought. Run automos with the lowpass cub below the Clementine64k cub(algebra cub). Only the null areas will have the lowpass cub show through.
This will fill in the null areas.
Did you apply a special "stretch" when running isis2raw?
cartrite

No not special: ==> linear

But I extended the range from 0.5 ...99.5 down to 0.1 ..99.5

I had read about this in the iris3 forum, where people also complained about the darkness of planets etc.

Fridger

[t00fri]

Here is another thought. Run automos with the lowpass cub below the Clementine64k cub(algebra cub). Only the null areas will have the lowpass cub show through.
This will fill in the null areas.

Yeah that sounds promising!

Fridger

[cartrite]

I tried the lowpass file created during the filter run under the algebra file with handmos and it didn't come out right. But still maybe running lowpass with a smaller boxcar that is not so blurry may still work.
I still see a lot of seams / blocks in the file. Looks better than before though.
There are options to remove null areas from the equations run by lowpass. Maybe that should be done. It looked like the null areas may have darkened some areas?
cartrite

[cartrite]

It seems that running lowpass with default settings will change a null pixel to the average value of the boxcar only if it is the center pixel. So running lowpass a few times with a small boxcar may fill in the null areas. But I wish the were a way to change the pixel without including the null values in the average. Probably the smallest boxcar should be at least a few pixels bigger than the largest null area.

Then that can be used underneath when handmos is run. Automos requires a mapping group so that will give an error. Also, when I tried it last night, I exported the algebra cub to raw binary and imported it again into isis to get it back to 8 bit. So that meant that I "stretched" it twice. The highpass and algebra cub files were 32 bit. The lowpass cub was always 8 bit. I'm not sure if this was required or not. I'm redoing it this morning because I messed up last night and only scaled the boxcar up by a factor of 4.
cartrite

[t00fri]

It seems that running lowpass with default settings will change a null pixel to the average value of the boxcar only if it is the center pixel. So running lowpass a few times with a small boxcar may fill in the null areas. But I wish the were a way to change the pixel without including the null values in the average. Probably the smallest boxcar should be at least a few pixels bigger than the largest null area.

Then that can be used underneath when handmos is run. Automos requires a mapping group so that will give an error. Also, when I tried it last night, I exported the algebra cub to raw binary and imported it again into isis to get it back to 8 bit. So that meant that I "stretched" it twice. The highpass and algebra cub files were 32 bit. The lowpass cub was always 8 bit. I'm not sure if this was required or not. I'm redoing it this morning because I messed up last night and only scaled the boxcar up by a factor of 4.
cartrite

Hi Steve,

something I noted: the 8k testfile does NOT have the 0 pixels marked as NULL=special pixels. So these will probably not be found in NULL pixel hunts! Marking NULL pixels is a special request one has to do during the conversion! You see than when using the special pixel tool of qview. No blue enhanced black bars. Clementine64k.cup has it correctly set.
If you got a 8bit bin file and convert it to cub with raw2isis, then you must set this option!

Fridger