Reprocessing Historical Images - II, Restoring images from antiquated and/or poor quality sources Options

[PDP8E]

Emily,

I'll give it a shot !! (and thanks for throwing down the velvet gauntlet)

(designing non-linear programs for noise reduction and focus is my preferred way to relax!)

Ted <wizard!> where would a humble programmer obtain the calibrated source images of Proteus?


stay tuned

[elakdawalla]

In the spirit of teaching a man to fish...

Voyager Image Catalog Query Form 2.0
Select "PROTEUS" from the dropdown menu in the Neptune targets line, and I recommend you select "detailed" from the "Listing type" dropdown menu under the Query Results header at the bottom of the page, and also ask to see preview gifs. I get 42 records returned; you'll need to go back to the query form and tell it to skip the first 32 records to see the image you're after. (Alternatively, here's a direct link to the file, which you can convert from PDS format using IMG2PNG or NASAView.)

These data are pretty raw. The rings node is working on developing calibrated and geometrically corrected Voyager data sets, but to date they have only done so for Uranus and Saturn, and even those data sets aren't quite through peer review yet.

--Emily

[tedstryk]

Emily,

I'll give it a shot !! (and thanks for throwing down the velvet gauntlet)

(designing non-linear programs for noise reduction and focus is my preferred way to relax!)

Ted <wizard!> where would a humble programmer obtain the calibrated source images of Proteus?


stay tuned

I am not sure what you are asking. If you are asking for mine, I will post it.

Edit: I have posted my version in my post after the next one.

[tedstryk]

Hi,

Just my small contribution to this topic. Here's a quick processing of the Proteus image. I was completelly conservative on appliying gausian blur filters to reduce noise and levels adjustment. There's some cliping of the darkest parts of the image in an atempt to reduce uncertain details at high noise levels.

The general details visible on Ted's images are present, and a hint of the finer ones.

I think that the issue here is the balance between further processing and the introduction of processing artifacts. It's a complicated decision and is related to image scale and sampling.

4throck, what I have basically done is combined that with the sharp high contrast features (such as the limb and terminator features) that did not require any blur to make clear.

[tedstryk]

Here is my calibrated base. Keep in mind that in this version, the sections of the Proteus frame that were not transmitted look awkward due to the fact that they were filled in with pure black on the PDS raw image.

[tedstryk]

Here is another image that is shown in a very odd form, the HST Venus image. First of all, it is given a garish false color. Second, its contrast is stretched beyond reason, creating a false crescent.



I have avoided colorizing the image, and left a more appropriate curve while enhancing the UV details.

[tasp]

If this suggestion involves much work, my apologies, but if it is just some click and drags (and no one has tried this yet) go for it;

would it help to verify an image processing technique to try it on, for instance, a Voyager Encedalus image, and then compare the enhancement to a subsequent Cassini image of the same area ??

If a particular technique makes particular details stand out in a Voyager image that are verified in a Cassini image, then perhaps if the same were done to, for example, a Voyager Uranus Puck image, then we might have some expectation the technique is worthwhile.

I might then suggest using the technique on some Cassini images . . .

(heh, heh)

[tedstryk]

The data issues with those are totally different. tasp, I find your snide "if it involves too much work" comment interesting....why don't you do it if it is so easy? There isn't a dataset out there (to my knowledge) like the Proteus image. So sharp, so close, and yet so underexposed. The fact that the major features all corresponded to the other images is good enough for me. I realize that noise is still a factor, but my position versus Tayfun Oner's is basically one of being willing to leave salt and peppering noise to cue a viewer that the picture is noisy or create a clean looking product that might lead an uninformed viewer to mistake noise for a real feature. When I am trying to make pretty pictures, the later is my choice. I would be much more careful if I was working on an image for scientific purposes, such as on my work with Ariel.

[tasp]

I have no way of knowing how much effort is involved in processing an image once a technique is developed. I have (don't laugh) HP image Zone on my PC and you just select a picture and click various settings on it and view the result. I am sure all of you are doing things much more advanced than that, and I have absolutely no idea how much effort is involved in restoring a particular picture once you have written some software to do it, nor could I have any idea unless specifically informed about a specific process.

I did not want to have anyone go to a lot of effort on one of my suggestions, but if it is easy to do, like I said, go for it. Don't anyone put themselves out on my account. I would never request someone put hours into trying out something I came up with, but if it s just a few seconds operation involving a click and drag, why not give it a try ?

Was not trying to be snide (just skirting topic of how dumb I am in regards to how computer image processing works) so apologies if that is how it came across.

[PDP8E]

Well....it has been about 2 weeks...two weddings, one funeral, one tropical storm, and the start of school again...but I did promise I would take a swing at the Proteus image.


First, my congrats go to Ted and the NASA people who have tried to work with this image. There are only a half dozen or so of actual DNs in the image. I took each DN level and processed it and then merged them. The eastern limb is a WAG ( wild a** guess) but there are faint impressions of a crater in the northwest and an elongated crater, centered in the due west bright area.

Warning: for obvious reasons, this is a highly processed image.....your mileage may vary...

(thanks to Emily for teaching an old angler to fish, and Ted for someone to aspire too!)

[tedstryk]

One thing that I wanted to emphasize. The thing that bothers me about the NASA version is that it is a quick-n-dirty from right after the encounter. There are better versions from the science teams of this and a lot of these images.

[jekbradbury]

I decided to take a look at the Proteus image as well. My goal when processing these is aesthetic in nature: create an image that looks like a moon, not a flat bumpy rectangular thing. The main difference between this image and better images is in the frequency histogram, aka the Fourier Transform. Correct that, and the image looks a lot more real. Therefore, the main operation I used was a low-pass filter to enhance the feeling of sphericalness and decrease noise, all while keeping enough small-scale details. Below is an animated gif showing three steps: a stretched, dereseaued, and deposterized (is that a word?) version of Ted's calibrated base, a sharpened and medium-pass enhanced version of the base, and the final product.

Below is an upscaled (SmartEdge algorithm) version.

[4th rock from th...]

Hi all,

I wrote a small program in Actionscript 3 (Flash) that performs linear interpolation to 8 bits on images with low bit levels originally. It's very basic and only works in one axis, but i can get around that simply rotating the original image 90º and averaging the results with the non rotated interpolation.
I tried it on the Voyager 2 Proteus image discussed in earlier posts. I must say I like the results.
The raw image was histogram stretched and cleaned of noise before interpolation. The interpolated image was then processed in Photoshop.

What I got is the image bellow.



Any ideas regarding other good images to test the program, besides the Viking sunset/sunrise images that I've already thought about?

[tedstryk]

That looks like a good result, but could you explain your use of "interpolated." Usually it is used to remove missing lines and pixels, but I am assuming this is used to reduce the "stair-stepping" effect. Am I correct?

[4th rock from th...]

"I am assuming this is used to reduce the "stair-stepping" effect."

Yes, that's it. I'll try to explain better:

Let's imagine that he have an image with pixel values, along a row, something like this:

100 110 110 110 120 120 110 100

What the program does is reading each pixel value and counting for how many pixels is that value the same as the previous pixel.
It's just like saying, in common language: there are 3 pixels in a row with a value of 110, and then the next pixel's value is 120.
Then the program divides the total pixel value difference (in this example 120-110=10 levels) by the amount of pixel the value stayed the same (3 pixels).
These 3 pixels with the same value are then replaced by a gradient (110 113 116).

So the resulting image data would be something like:

100 110 113 116 120 120 110 100


Hope this makes things a little clearer. I've tried it on the viking images, and it does give a smooth sky, but noise creates artifacts. So i need more time to optimize the code and try to ignore level changes due to isolated pixels.