Printable Version of Topic

Posted by: Harry Oct 17 2005, 10:22 AM

The following images are for Tempel 1 originally taken by NASA's probe (left) and its de-convoluted image (right). For details of the technique used for that de-convolution, please visit;

http://139.134.5.123/tiddler2/c22508/focus.htm

Posted by: tfisher Oct 17 2005, 01:28 PM

I took a look at the site you link to; is that program yours? I hope some critique won't come across as offensive. It looks like its claims are rather exaggerated. It just convolves the image with a sharpening kernel, with some parameters on that kernel. The main claim to fame of the parameters it provides seems to be iteration of a small sharpening kernel (where by small, I mean a pixel is affected by pixels from within a small radius). Iterating a small kernel gives some larger kernel; unfortunately this larger kernel will not be the best kernel of that size for most sharpening uses. For instance, you might hope your large sharpening kernel would be pretty much radially symmetric. Iterating a small kernel gives a bad approximation of this.

The Tempel 1 image you post does show some details better than the original (good!) but also has some classical signs of oversharpening like JPEG artifacts being exaggerated and some horizontal lines appearing near the top and bottom edges.

Posted by: ElkGroveDan Oct 17 2005, 03:39 PM

OK that is cool Harry. I can use a quickie utility like that. Is it your program?

As with all such tools I prefer to use the revised product as part of a multi-layer image with various transparency levels. Using different algorithims in different layers has proven successful for me. One more tool in the bag of tricks is always appreciated.

Posted by: Harry Oct 18 2005, 09:58 AM

To: tfisher

Yes, it's very important to determine the extent of kernel properly. But it must follow heuristic manner. For, if we were to know it previously, we would have corrected the focus before taking the photo!

>Iterating a small kernel gives a bad approximation of this.

That's true. Practically when choosing small kernel on my software (i.e. "Focus Depth" ~ 1), the effect of de-convolution is almost same as that obtained from usual sharpening process.

As you've indicated, some artifacts may appear after applying the software. Those seem to happen less frequently when the quality of original picture is better. If the picture is already processed (s.t. enlargement, filtration, etc.), the artifacts on the picture after applying my software will appear much more frequently.


To: ElkGroveDan

>Is it your program?

Yes, it is. I've developed this software for just correcting the focus on the image, not comparing the result with other software. Namely I haven't compared the sharpened image obtained from my software with that of other sharpening software by myself! If you get better outcome using some other software, could you let me know?

Posted by: deglr6328 Oct 18 2005, 10:11 PM

Nice work. It really brings out considerable detail in the original. But can it actually be called a deconvolution without direct knowledge of the point spread function of the telescope on DI...? Or am I missing something?

Posted by: Harry Oct 19 2005, 12:17 PM

As you've known, the form of point spread function depends on the optical system on which the photo is taken. Since I've developed this software for multi-purpose (i.e. not only for blurred planetary images but also for blurred microscopic images, etc.), the point spread function applied in the software has the simplest form, that is, the point of light is supposed to spread uniformly on the disc. (Do you think it's too simple?)

The following images are for Jupiter's satellite Thebe originally taken by Galileo probe (left) and its de-convoluted image (right). In this time the area of disc mentioned above is set four times greater than that of Tempel 1, since the original image of Thebe seems more blurred than that of Tempel 1.

Posted by: tfisher Oct 19 2005, 04:24 PM

The origin of the blur in the Thebe image you posted is not in the telescope system, but rather in the imaging team's reprocessing of a tiny image. Here is the original image (at full resolution and with a 5x pixel zoom so you can see it), as available http://pdsimg.jpl.nasa.gov/cgi-bin/Nav/navigator/browse_page?imagenum=1.



As you can see, there isn't that much information available in the picture to begin with -- somewhere around 100 non-black pixels. Sure, maybe you can improve on the original reprocessing somewhat to make a more visually pleasing enlargement, but trying to pull more information out than was in the original data is folly.

Posted by: Harry Oct 19 2005, 11:37 PM

Oh, I didn't know the image of Thebe which I've processed was not raw image! By the way, if available, could you upload the raw image of Thebe? For I could not access the link you've shown in the last post. (Or is the attached image in your last post a genuine raw image? I'm afraid of that image is too small to process with my software...)

Posted by: tedstryk Oct 20 2005, 12:03 AM

Yes, the image he processed is original. From the result, it seems that the software is just sharping the image, not doing true deconvolution. It is sort of like when I produce what I call "faux-super resolution images." It improves the visibility of already visible features, but, unlike true super-resolution imagery, doesn't yield true new detail.

Posted by: Decepticon Oct 20 2005, 02:02 AM

Could this work with the recent HST images of Ceres?

Posted by: tfisher Oct 20 2005, 04:46 AM

Sorry, I didn't notice the link I posted was to a cgi results page. Go to http://pdsimg.jpl.nasa.gov/cgi-bin/Nav/GLL_search.pl?TARGET_NAME=THEBE and click submit.

The image I posted is a direct crop from the raw image. So yes, it is really that small to begin with!

The good news is, if you browse through that data set there are more views of Thebe (and the other moons, etc) that haven't had any public exposure, as they only make press images of the best shots. (Yes, this image was one of the top three best shots ever taken of Thebe!) So you can find images to play with that pretty much only the original mission scientists ever looked at.

Posted by: Harry Oct 20 2005, 09:03 AM

Thank you for letting know me the site. It's a very interesting site. I de-convoluted some of those pictures in that site. In those pictures the left side is for original images and the right side is for the images after de-convolution. As you'll see, the de-convolution software is not effective when the original picture is composed of small number of pixels.

Posted by: Harry Oct 20 2005, 09:11 AM

Probably... but I'm not sure.

Posted by: Harry Oct 20 2005, 09:13 AM

OK. If I found a good (and blurred) image of Ceres, I'll try to de-convolute it.

Posted by: tedstryk Oct 20 2005, 09:33 AM

Here are all the resolved views of Thebe (shown at original size) we have....very slim pickings.

Posted by: Harry Oct 20 2005, 09:36 AM

And this is the de-convoluted image obtained from the image you've posted.

Posted by: Harry Oct 21 2005, 01:54 AM

The following pictures are the images of Ceres originally taken by HST (left column) and their de-convoluted images (right column). But please mind those "de-convoluted" images may not be correct but have just exaggerated contrasts, since those pictures' resolutions are low and I've not seen any blurred pictures taken by HST after implementing COSTAR.

Posted by: Decepticon Oct 21 2005, 12:18 PM

Neat!

Posted by: Harry Oct 22 2005, 04:50 AM

The left side is the image of Pandora originally taken by Cassini probe and the right side is its de-convoluted image.

Posted by: Harry Oct 23 2005, 04:32 AM

The left side is the image of Dactyl originally taken by Galileo probe and the right side is its de-convoluted image.

Posted by: deglr6328 Oct 23 2005, 06:28 AM

I think you may have re-discovered the p-shop unsharp mask algorithm!
Original on left, your image middle, p-shop unsharped original on right.

Posted by: Harry Oct 24 2005, 04:23 AM

I didn't know p-shop unsharp mask. I'll check it.
I should have noted the parameters used for de-convolution. The following pictures are (from left:) the original image of Dactyl, de-convoluted image A (focus depth:= 1.8, iteration:= 5), B (focus depth:= 2.2, iteration:= 5), and C (focus depth:= 2.6, iteration:= 5)
Also I attached the image of Janus taken by Cassini probe and its de-convoluted image (focus depth:= 2.2, iteration:= 5)

Posted by: tedstryk Oct 24 2005, 02:00 PM

The problem is that the other than the HST pre-repair Jupiter image and the Tempel 1 image, these images are not out of focus - they are simply limited by pixel size. So while the sharpening is nice, it isn't pulling out any new information.

Posted by: ugordan Oct 24 2005, 02:15 PM

Quite true. "Deconvolving" obviously bilinearly or bicubically enlarged images has no point. This thing reminds me of a "wonder" procedure some folks applied on low quality encoded mp3 files. They used to decode them into wave audio and then re-code into a much higher bitrate, thinking the quality's going to improve. Then, by the power of auto-suggestion the file really does sound better.
Seriously, though, I believe all that is a result of too many Hollywood movies. How many times have you seen forensics zooming and magically "enhancing" analog or digital images and virtually getting whatever amount of detail it's needed by the plot script...

You can't pull a rabbit out of a hat if you don't have the hat!

Posted by: Harry Oct 24 2005, 11:21 PM

The image of Vesta taken by HST (left) and its de-convoluted image (right / Software: http://139.134.5.123/tiddler2/c22508/focus.htm, Focus Depth:= 4.3, Iteration:= 8)

Posted by: tedstryk Oct 25 2005, 01:40 AM

Note the bright right edge. This is clearly sharpening, not focusing.

Posted by: Decepticon Oct 25 2005, 01:58 AM

You can clearly see a crater at the upper left corner.

Posted by: Harry Oct 25 2005, 03:02 PM

Thank you. The following pictures are the original image of Vesta (left), its de-convoluted image (middle / focus depth:=3.8, iteration:=8) and another de-convoluted image (right / focus depth:=4.3, iteration:=8) The right image is the same as the de-convoluted image in my earlier post.
Also I attached the image of Prometheus and its de-convoluted image (focus depth:=1.8, iteration:=5).
By the way, it is really hard work to find any blurred images among the published images taken by Cassini probe.

Posted by: tedstryk Oct 25 2005, 05:53 PM

"By the way, it is really hard work to find any blurred images among the published images taken by Cassini probe."

You would think that they had the camera set at infinity focus or something!

Posted by: Harry Oct 26 2005, 11:33 AM

The image of Annefrank taken by Stardust space probe (left) and its de-convoluted image (right / Software: http://139.134.5.123/tiddler2/c22508/focus.htm, Focus Depth:= 2.4, Iteration:= 5)

Posted by: Decepticon Oct 26 2005, 12:13 PM

Now thats cool! ^^

Posted by: tedstryk Oct 26 2005, 01:36 PM

That image is already over-processed!

Posted by: Harry Oct 26 2005, 11:42 PM

Thank you. Well, I think I did Focus Correct for all blurred images which I could get. I heard NASA sends the probe to Ceres and Vesta in near future. I look forward to looking at the images sent from that probe.

Posted by: Harry Nov 21 2005, 10:57 AM

The following pictures are the image of Calypso taken by Cassini probe (left) and its deconvoluted image (right) processed by http://139.134.5.123/tiddler2/c22508/focus.htm (focus depth: 3.4, iterations: 9).

By the way through the discussions on this forum, I knew the Dawn project might become pending. Oh...

Posted by: ugordan Nov 21 2005, 11:57 AM

You do realize that your method doesn't bring out any real details from these images, don't you? The reason is that there are no further details beyond the resolution capability of the camera. All you accomplish is sharpening the noise or in this case noise coupled with bilinear enlargment artifacts.
Processing images after they have been taken at a low resolution and magnified is not the same as deconvolving images that are blurred by the optics, but still are captured at the nominal resolution. In the first case detail is permanently lost, while in the latter case detail was present but was blurred beyond recognition.

If you had two distinct Cassini Calypso images taken immediately one after the other and applied your sharpening, you'd quickly find out the "features" in the two images would not correspond to one another, a clear sign of them being synthetic.

Posted by: Phil Stooke Nov 21 2005, 03:42 PM

ugordan said:

"If you had two distinct Cassini Calypso images taken immediately one after the other and applied your sharpening, you'd quickly find out the "features" in the two images would not correspond to one another, a clear sign of them being synthetic."

Very true. But if you applied this process to both of them and then merged the results, the noise would be reduced and the real features would stand out better. But basically of course ugordan is right, you can't create new details!

Phil

Posted by: Harry Nov 23 2005, 12:01 PM

My software corrects the blurry on the image but can NOT improve the resolution which the original image has. It is a difficult question to answer whether the blurry appeared on the image has been caused from out of focus or low resolution since the blurry may be caused from the movement of object during the exposure even if the camera has the right focus and sufficient resolution.

The following images are the comet 67P/Churyumov-Gerasimenko originally taken by HST and its de-convoluted image processed by http://139.134.5.123/tiddler2/c22508/focus.htm (focus depth: 3, iterations: 7).

Posted by: tedstryk Nov 23 2005, 12:50 PM

Nice enhancement of a shape model!

Posted by: ugordan Nov 23 2005, 12:58 PM

Ooooo, BAD!!!

Posted by: Phil Stooke Nov 23 2005, 10:46 PM

I wouldn't personally place too much faith in the original shape model here. Its alarming degree of symmetry is a sure sign that the model is straining to produce a result at the limits of the data. If not beyond.

But Harry, this is a rendered image of a mathematical model, not the original HST image. Your effort would be more useful if you applied it to the original image rather than this. But it's good that you are experimenting with these things.


Phil

Posted by: Harry Nov 24 2005, 01:33 PM

Thank you for your kind suggestion. It may be meaningless to de-convolute the image which has been already reprocessed. However I thought occasionally it might reveal some unknown features of original image.

Following your suggestion, I de-convoluted the genuine image of Vesta taken by HST. In the figure the left side is for the original image and the right side is for the de-convoluted one (Software: http://139.134.5.123/tiddler2/c22508/focus.htm, parameters: focus depth:= 4, iterations:=9).

Posted by: Harry Nov 25 2005, 12:42 PM

The image of Vesta taken by HST (left) and its de-convoluted image (right) processed by http://139.134.5.123/tiddler2/c22508/focus.htm (focus depth:=4, iterations:=9)

Posted by: Harry Nov 26 2005, 10:41 AM

The image of Vesta taken by HST (left) and its de-convoluted image (right) processed by http://139.134.5.123/tiddler2/c22508/focus.htm (focus depth:=4, iterations:=9)

Posted by: BruceMoomaw Nov 26 2005, 10:49 AM

Once again: while I hate to say it, to the extent that these Huble images are fuzzy due to a simple shortage of pixels rather than to actual defocusing of the telescope's mirror, you're engaged in a fool's errand -- your program will just be synthesizing nonexistent details. And the Hubble images of little tiny distant bodies like Vesta and Pluto are indeed seriously limited in their pixel width -- those smooth-looking views of them that are published are misleading, being themselves the product of computer programs designed to make the original images look much less "grainy".

Posted by: tedstryk Nov 26 2005, 03:44 PM

Bruce is right. Other than the Deep Impact HRI, pre-repair Hubble, and a Lunar Orbiter camera or two, I can't think of any truly out of focus camera. You are simply producing arifacts.

Posted by: BruceMoomaw Nov 26 2005, 09:37 PM

You know, I hadn't thought of that, but I suppose deconvolution could be applied fairly easily to those Lunar Orbiter 1 photos from its high-resolution lens that were smeared because the orbital image-motion compensation system was interfered with by that lens' shutter. The blurring in that case was just in one direction. (But then, they only took a few photos with it before sensibly using the medium-resolution lens for the rest of the mission.)

Somewhere, on one of those big pictorial brochures that the New Horizons team has released, there's an "original" version of Hubble's Pluto photos that show just how jagged-looking and "Cubist" those photos are before they're deliberately smoothed-up for public consumption. The situation won't be as bad for Hubble's photos of Ceres and Vesta; but still, as I say, deconvolution programs simply can't and don't do any good at all for any of Hubble's photos after its mirror was corrected. You're wasting your time on those.

Posted by: edstrick Nov 27 2005, 02:53 AM

The problem with trying to deconvolve the image-motion-smeared Lunar Orbiter 1 pics is that the smear is large.. tens to hundreds of pixels, and the data is horribly non-quantitative with horrendous levels of readout and analog-transmission and recording artifacts. You'd end up enhancing artifacts, not restoring details, I am pretty well convinced.

More useful to try to restore are the Orbiter IV images that were literally fogged by condensation on the camera box window when they had problems with the aperture door sticking open early in the mission. There was variable amounts of patchy condensation on the window surface and many images from the first quarter to third of the mission were SEVERELY degraded.... basically everything from Tranquilitatis or Serenetatitis to the east limb and a bit beyond.

Remember, the film was developed on the spacecraft by a "wet bimat" process... sort of like the film and polaroid processing that produced b&W negatives... so there had to be moisture in the camera and film cannister.

Posted by: tedstryk Nov 27 2005, 03:10 AM

It depends on the type of source. For point sources, and other high contrast sources, it works pretty well. But for extended, low contrast sources, it can do little. Also, it works better with WFPC (those Pluto images are FOC), because due to its lower resolution, the problems don't spread light over as many pixels.
This is a commonly seen HST Mars image from 1993 (this is my version).


However, much of the bluring in this deconvolved image is due to the fact that it is enlarged. Here is the image at original resolution:



Also, a lot of difficulty in processing this is that the rotation of Mars between images is TERRIBLE.

Posted by: BruceMoomaw Nov 27 2005, 06:26 AM

An original Hubble image of Vesta showing the pixel graininess is at http://hubblesite.org/newscenter/newsdesk/archive/releases/1997/27/image/d . It's obviously much better than Hubble's best Pluto views -- which can be found on page 15 of http://pluto.jhuapl.edu/overview/deis/presentations/PublicMeetingPresentations.pdf -- but you can still see that the value of deconvolution would be very small.

Posted by: Harry Nov 27 2005, 12:27 PM

Oh, please don't think it too seriously. I've not intended to report some scientifically rigid results here, but just have proposed a possible way to guess the features of Vesta...

Posted by: BruceMoomaw Nov 28 2005, 02:19 AM

OK, but if it doesn't actually give us any information on real features it is -- to put it mildly -- an idle hobby.

Posted by: mike Nov 28 2005, 03:01 AM

Just say you're making art..

Posted by: Harry Nov 28 2005, 10:47 AM

The image of Vesta taken by HST (left) and its de-convoluted image (right) processed by http://139.134.5.123/tiddler2/c22508/focus.htm (focus depth:=4, iterations:=9)

Posted by: The Messenger Nov 28 2005, 04:30 PM

We used to get in this argument a lot, when evaluating X-ray deconvolution schemes. If the material being evaluated was well known and characterized, there are sub-pixel routines that will truly improve resolution. However, if there are unknown contaminates, and expecially contaminates of the same size as the pixels, these routines actually remove information.

Posted by: Harry Nov 29 2005, 11:59 AM

Although it is not the picture for asteroids, I attached the de-convoluted image processed by http://139.134.5.123/tiddler2/c22508/focus.htm as a reference. (left: original image of Supernova 1987A, right: its de-convoluted image)

Posted by: Harry Nov 30 2005, 10:46 AM

The image of Callisto taken by HST (left) and its de-convoluted image (right) processed by http://139.134.5.123/tiddler2/c22508/focus.htm (focus depth:=2.4, iterations:=5)

Posted by: Harry Dec 1 2005, 02:56 PM

The image of Io taken by HST (left) and its de-convoluted image (right) processed by http://139.134.5.123/tiddler2/c22508/focus.htm (focus depth:=2, iterations:=7)

Posted by: ugordan Dec 1 2005, 03:02 PM

This would be an excellent time for someone with the know-how to render an Io view of the same spatial resolution and central coordinates from the Galileo data so we could actually compare how well the features match up.
Any volunteers?

Posted by: JRehling Dec 1 2005, 07:04 PM

If valiadating the approach is the point, an easier exercise would be to downsample some images, let him run his magic on them, and see if the details in his results correlate with the original. For example, take a 1000x1000 image of the Moon, shrink it to 100x100, then re-enlarge the shrunken version to 1000x1000, and see if Harry's algorithm recreates real features smaller than 5 pixels in the original. If not, then the details should be considered fiction.

That would be easier than trying to project Io imagery. The HST images of the Galileans, if the unusual colors are any indication, did not use the same (or even similar) filters as Galileo.

Posted by: Airbag Dec 2 2005, 03:02 AM

No, that is not the same at all - you can't create information when it is not there to begin with.

Deconvolution works when trying to correct an image that has undergone convolution, e.g. because of the effects of diffraction limited optics. Deconvolution will then attempt the reverse mathematical process to "undo" the original convolution. Of course, in all image processing operations, some information is lost - it is just a matter of what kind of information you want to optimise at the expense of others.

Note that deconvolution only really works if you know the specifics of the original convolution process, e.g. the spherical abberation of the HST's main mirror. If you just apply some random deconvolution until you start seeing things, you probably just are.

Airbag

Posted by: ugordan Dec 2 2005, 08:32 AM

That was precisely what I was trying to get at. To prove to Harry once and for all that the details he brings out of the images are processing artifacts, nothing else.

JRehling, you're absolutely right, your approach would be much simpler to do than what I proposed. So simple in fact that it makes me wonder why I haven't thought of it before

Posted by: Toma B Dec 2 2005, 09:26 AM

In other words there is NO DECONVOLUTION...It's just what's in Adobe Photoshop called Brightness/Contarast/Sharpen etc...
Sorry Harry that's just not it...
Nice little pictures thou...

Posted by: Harry Dec 2 2005, 10:29 AM

Honestly saying, I'm not sure whether the blurry appeared on the image was owing to the reprocessing or the optical system on HST.

As the reference, I attached again the previous images (their colors have been changed) and the image of Io taken by Galileo probe.

Posted by: ugordan Dec 2 2005, 10:37 AM

Those images of the four Galilean satellites were taken some time before Galileo arrival, IIRC, and *after* the optics on the HST have been fixed. It's simply because the resolution of the satellites isn't all that great and the images were *magnified*, hence the blurry look. No optics issues.

Posted by: tedstryk Dec 2 2005, 02:16 PM

Well, in the case of HRI on Deep Impact, where it is a matter of simply being out of focus (as opposed to spherical aberation on HST, where some light is brought to a focus, while other light is out of focus to varying degrees). Is that if you have some point source images, you can figure out the point spread, which will allow you to proceed. But as for pulling sub pixel information out of single images, it simply can't be done - you can't create information that is not in the image.

Super-resolution is somewhat effective because it utilizes multiple images, taking advantage of the fact that their pixels don't overlap exactly (if they do, it doesn't work).

Posted by: tedstryk Dec 2 2005, 02:18 PM

Harry, Ugordan is right. That image of Io is not blurry. It is just greatly enlarged.

Posted by: Harry Dec 3 2005, 10:53 AM

As you all imagine, it's indeed hard task to find any blurred images which are not reprocessed among published astronomical photos, for any blurred images are reprocessed, or simply are not published.

The following images are for Vesta taken by HST and its de-convoluted image processed by Focus Corrector (focus depth:=4, iterations:=9).

Posted by: deglr6328 Dec 3 2005, 07:01 PM

But that Vesta image IS processed. If it werent it would look like this or simillar.

Posted by: tedstryk Dec 3 2005, 07:50 PM

In the case of most missions, they are imaging targets that are at an infinity focus, and their focus doesn't change. Real blurring is from vibrations and motion smear. And that Vesta image is enlarged.

Posted by: Harry Dec 4 2005, 01:12 PM

Oh, is it a moon or asteroid? I tried to de-convolute that image with Focus Corrector. In the following figures the left side is the picture you've posted (with downscaling to 1/2) and the right side is its de-convoluted image.

Posted by: Harry Dec 4 2005, 01:44 PM

And I applied median filter to the image you've posted. The following pictures are the image after applying median filter (left) and its de-convoluted image processed by Focus Corrector (right).

Posted by: Harry Dec 5 2005, 10:04 AM

The image of Vesta taken by HST (left) and its de-convoluted image (right) processed by http://139.134.5.123/tiddler2/c22508/focus.htm (focus depth:=4, iterations:=9)

Posted by: Harry Dec 6 2005, 01:30 PM

The image of Vesta taken by HST (left) and its de-convoluted image (right) processed by http://139.134.5.123/tiddler2/c22508/focus.htm (focus depth:=4, iterations:=9)

Posted by: Rob Pinnegar Dec 6 2005, 01:38 PM

Those are some nice pictures, Harry. Deconvolution is a bit of a tricky process --we also use it in geophysics, so I know a bit about it -- but if done right it can produce good results.

By the way, although the process is called "deconvolution", when it has been applied to an image we say that the image has been "deconvolved" rather than "de-convoluted". That's just a bit of jargon. It does sometimes confuse people.

Posted by: Harry Dec 7 2005, 01:57 PM

Thank you. In a discussion I heard some person used "deconvolve" but other person used "deconvolute". In the field of geophysics, the verb "deconvolve" is commonly used?

The following images are for Vesta taken by HST (left) and its "deconvolved" image (right) processed by Focus Corrector (focus depth:=4, iterations:=9)

Posted by: Harry Dec 9 2005, 12:02 PM

The image of HH34 taken by HST (left) and its de-convoluted image (right) processed by http://139.134.5.123/tiddler2/c22508/focus.htm (focus depth:=1, iterations:=5)

Posted by: Harry Dec 10 2005, 12:22 PM

The image of Vesta taken by HST (left) and its de-convoluted image (right) processed by http://139.134.5.123/tiddler2/c22508/focus.htm (focus depth:=4, iterations:=9)

Posted by: Harry Dec 12 2005, 01:20 PM

The image of centre of radio galaxy Centaurus A taken by VLT (left) and its de-convoluted image (right) processed by http://139.134.5.123/tiddler2/c22508/focus.htm (focus depth:=2, iterations:=6)

Posted by: Harry Dec 14 2005, 01:47 PM

The image of Vesta taken by HST (left) and its de-convoluted image (right) processed by http://139.134.5.123/tiddler2/c22508/focus.htm (focus depth:=4, iterations:=9)

Posted by: Harry Dec 15 2005, 01:45 PM

The image of Vesta taken by HST (left) and its de-convoluted image (right) processed by http://139.134.5.123/tiddler2/c22508/focus.htm (focus depth:=4, iterations:=9)

Posted by: Harry Dec 17 2005, 02:15 PM

The image of Vesta taken by HST (left) and its de-convoluted image (right) processed by http://139.134.5.123/tiddler2/c22508/focus.htm (focus depth:=4, iterations:=9)

Posted by: Decepticon Dec 17 2005, 05:53 PM

Harry a request SVP!?

What can been done with this Ceres picture?

http://www.astronomy.com/asy/objects/images/ceres_800.jpg

Posted by: Harry Dec 18 2005, 12:10 PM

Ok. I tried to de-convolute the image after downscaling it. In the following images, the left side is the original image and the right side is the de-convoluted one. As you'll see the edge of each pixel is enhanced. Personally I feel the features of Ceres is still enigmatic. I hope someone brings the picture of Ceres with better resolution in future...

Posted by: Harry Dec 19 2005, 02:17 PM

The image of Vesta taken by HST (left) and its de-convoluted image (right) processed by http://139.134.5.123/tiddler2/c22508/focus.htm (focus depth:=4, iterations:=9)

Posted by: Harry Dec 21 2005, 01:17 PM

The image of Vesta taken by HST (left) and its de-convoluted image (right) processed by http://139.134.5.123/tiddler2/c22508/focus.htm (focus depth:=3.8, iterations:=11)

Posted by: Bob Shaw Dec 21 2005, 01:35 PM

Harry:

I thought of a way to demonstrate to everyone whether or not your techniques are really pulling hidden data out of images, or are instead just imposing artefacts onto the original data.

It's simple - test shots.

We need a series of images of known objects, such as the Moon, or famous landmarks (both natural and man-made). These should be 'out-of-focus' (NOT blurred in Photoshop or whatever, goodness knows what that would do to the test) and should be matched by, er, in-focus shots of the same things taken at the same time.

You apply your techniques to the blurry jobs, then we all look at the effects - with the in-focus versions as a sanity check.

That should settle matters!

Bob Shaw

Posted by: Harry Dec 23 2005, 11:38 AM

Thank you for your suggestion. At present I don't have any blurred published images of Moon. I'll look for some photos suitable for that purpose. Meanwhile, please check the following images of Tempel 1. The left is the original image taken by the Kamikaze probe with magnified twice, the middle is its de-convoluted image (Software: http://139.134.5.123/tiddler2/c22508/focus.htm, focus depth:=2.6, iterations:=7), and the right is another original image taken at the position closer to the surface (refer to http://photojournal.jpl.nasa.gov/catalog/PIA02128.)

Posted by: ugordan Dec 23 2005, 11:46 AM

After all the recent arguments, you still insist on deconvolving magnified images. Don't you get it there's nothing to deconvolve in a magnified image?
I see Ted Stryk did some work on Hubble's pre-repair images of Mars in http://www.unmannedspaceflight.com/index.php?showtopic=1790&view=findpost&p=30208, why don't you also get those raw images and try your luck on them, for a change?

Posted by: Rob Pinnegar Dec 24 2005, 02:42 AM

Yeah. Although these deconvolved images _are_ nice to look at, I did notice recently that a disproportionate number of them (especially the Vesta series) seem to feature a dark "ring" just inside the boundary of the asteroid.

There's no way something like that could appear at all those different viewing angles. It's got to be an artifact of the technique.

Harry's approach may have some similarities with edge detection perhaps? Anyways, as is the case with a lot of image processing methods, the danger seems to be that, if you keep tweaking the parameters patiently enough, it might just show you what you want to see. That can spell trouble.

Posted by: edstrick Dec 24 2005, 09:52 AM

My impression is that this sharpening algorithm that's under discussion is a quite effective adaptive sharpening filter. While it does bring out nasty jpg artifacts and other problems in degraded images, with images that are "capable of being helped", it usefully boosts fine detail with a relative absence of ringing or other artifacts.

The dark ring around the boundary of the asteroid images is clearly an artifact; I suspect it results largely from the space background outside the asteroid's limb having been set to a uniform black, rather than being real data, with the result that the enhancement goes crazy beyond the edge of real data.

Posted by: Harry Dec 24 2005, 02:56 PM

"The dark ring around the boundary of the asteroid images is clearly an artifact; I suspect it results largely from the space background outside the asteroid's limb having been set to a uniform black, rather than being real data, with the result that the enhancement goes crazy beyond the edge of real data."

I hope so. However, perhaps it might have been caused from the accumulation of fractional errors. I must check whether there are possible bugs in the software or not...

I attached the image of Metis taken by Galileo probe (left) and its de-convoluted image (right / software: Focus Corrector, focus depth:= 3.6, iterations:=12).

Merry Christmas!

Posted by: Rob Pinnegar Dec 24 2005, 05:38 PM

I don't know about that, Ed. Something I just now noticed is that some of the Vesta images also have a faint (but visible) _bright_ ring just outside the asteroid's "boundaries". See 'em? (Besides that, the space background of the real data would be pretty close to a uniform black, one would think.)

Admittedly, it has been a while since I read up on the mathematics of deconvolution, but my gut feeling is that this could be a sidelobe effect. That's just a guess, though.

Posted by: Harry Dec 27 2005, 11:21 AM

The image of Mira taken by HST (left) and its de-convoluted image (right) processed by http://139.134.5.123/tiddler2/c22508/focus.htm (focus depth:=4, iterations:=10)

Posted by: edstrick Dec 27 2005, 12:15 PM

The image of Metis, I can believe those are mostly real details that are bing enhanced. for Mira, my image-processing-instincts say mostly not.

Posted by: mike Dec 27 2005, 05:40 PM

Yeah, but it looks cool.

Posted by: The Messenger Dec 27 2005, 07:11 PM

You're morphing it into a waffle.

Posted by: djellison Dec 27 2005, 07:23 PM

Yup - all it's doing is putting an echo of the dark area into the middle, and an echo of the light area outside it. It's certainly not producing any real features.

Doug

Posted by: BruceMoomaw Dec 27 2005, 10:41 PM

Waffles In Space! Miss Piggy would be delighted.

Posted by: Bob Shaw Dec 28 2005, 12:41 AM

So let's *test* the process with some known targets!

Bob Shaw

Posted by: Harry Dec 29 2005, 12:35 PM

Perhaps Mira is made of waffles. Since the universe is enough vast, there may be a star made of Turkish Delights, too.

Posted by: Harry Dec 31 2005, 11:52 AM

"The image of Metis, I can believe those are mostly real details that are bing enhanced. for Mira, my image-processing-instincts say mostly not."

The image of Mira may be just in focus. My software seems not work well when the original image has poor resolutions.

"So let's *test* the process with some known targets!"

Sorry, for I could not find appropriate example so far...

The following images are for Epimetheus taken by Cassini (left) and its de-convoluted image (middle / software: Focus Corrector, focus depth:= 1.8, iterations:= 8). As the reference I attached another image of Epimetheus taken at the position closer to the surface (right).

Posted by: Harry Jan 8 2006, 01:08 PM

The image of Vesta taken by HST (left) and its de-convoluted image (right) processed by http://139.134.5.123/tiddler2/c22508/focus.htm (focus depth:=4, iterations:=9)

Posted by: Harry May 23 2006, 02:11 PM

The image of Comet 73P/Schwassmann-Wachmann 3 (fragment "B") taken by HST (left) and its de-convoluted image (right) processed by http://139.134.5.123/tiddler2/c22508/focus.htm (focus depth:=2, iterations:=8)

Posted by: DonPMitchell May 24 2006, 05:47 AM

Not to keep nit picking, but "deconvolve" does not mean "sharpen", it is something different from what your program is doing.

In Fourier-transform space, convolution is equivalent to multiplying the image spectrum by a function f. Deconvolution tries to undo this by multiplying the spectrum by 1/f. It's a risky operation, because it can amplify noise. Norbert Wiener derived an optimal filter to bring out the most information given a certain noise level, called a Wiener filter, basically multiplication by 1/(f + c) for a magic value of c. Anyway, it is not just a sharpening kernel.

Photoshop CS2 has a filter called "Smart Sharpen", which I believe is true deconvolution. It seems to do an amazing job once you learn how to fiddle its parameters.

Posted by: DonPMitchell May 24 2006, 06:07 AM

This got me curious, so I fiddled with this image of Tempel. First, fetch the original TIFF (not JPG) image from NASA. Note that it is 250 x 250 pixels, so undo their pixel-replication by resixing it from 500 to 250 with a "nearest-pixel" option in Photoshop. If you don't do that, all bets are off!

Then I ran it though Smart Sharpen deconvolution, with a radius of 1.5. More than that and the image rings, which indicates you made the radius too big. Here is the original and the deconvolved images, expanded again to 500 with a windowed sinc filter:

[attachment=5837:attachment]

I think the deconconved image is quite a bit cleaner than the sharpened image, although that may have jpeg artifacts amplified, or perhaps it was not downsized to 250 first?

Posted by: Bob Shaw May 24 2006, 08:13 AM

Don:

Very nice! The 'new' detail is indeed very credible...

Bob Shaw

Posted by: DonPMitchell May 24 2006, 05:07 PM

In the technical papers about this camera system, there will probably be (should be!) a measurement of the aperture (convolution blur) function of the camera. So ideally, you would try to invert that.

Detail is revealed by amplifying higher frequencies, but this also amplifies the noise in the image, so you have to take care or you end up finding UFO bases in NASA photos (a favorite pass time for some folks)!

Posted by: Comga May 26 2006, 05:23 AM

Indeed, the Deep Impact team has many star images to use as PSF models for deconvolution. They have this for the monochromatic images and through each filter. They also have computer models of the optical system which accuratly mimic the condition of the mirrors. With these, the don't need to do a "blind" sharpening but can do a true deconvolution, using one of several algorithms.

Posted by: Harry May 26 2006, 07:00 AM

My method is different from Fourier transform. As you know, the output image g(x) is expressed as,

g(x) = f*n(x)

where the extent of n(x) corresponds to the deviation from the focal point (if g(x) is just in focus, then n(x) = ((delta))(x).) After applying Fourier transform to the above equation, we get;

G(u) = F(u)N(u)

hence,

F(u) = G(u)/N(u)

Therefore, by applying inverse Fourier transform to G(u)/N(u), we can get the image data f(x) before convolution. While my method is to solve the first equation: g(x) = f*n(x) directly. Firstly that equation is discretized as,

g(j) = ((sigma))_k f(j-k)n(k)

It is expressed with matrix-vector form as,

g = Af

hence,

f = A^(-1)g

The problem is how to calculate f(x) from the above equation. You may imagine to apply Gauss-Seidel method for it. But it can not be applicable in this case because A doesn't satisfy the condition in which the iteration by Gauss-Seidel method converges. Regarding the method I took, please refer to: http://139.134.5.123/tiddler2/c22508/iteration.htm

The attached images are the comparison of the original image (left)/the image you've got by Photoshop (middle-left)/the image obtained by Focus Corrector (middle-right)/another genuine image taken at the position closer to the surface of Tempel 1 (right). The parameters of Focus Corrector used for that image are:

Focus Depth = 1.7
Iterations = 7

Posted by: Harry May 27 2006, 11:20 AM

The image of Supernova 1987A taken by HST (left) and its de-convoluted image (right) processed by http://139.134.5.123/tiddler2/c22508/focus.htm (focus depth:=1.7, iterations:=7)