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

[tedstryk]

I figure that it is time for another thread like this. I still find it astonishing to see the versions of a lot of images that are reproduced over and over again. For example, this is the last mosaic of Triton taken before the close encounter began. The version on the Planetary Photojournal is on the left, my version on the right. Clearly, this image was produced as part of the "instant science" campaign. They did a superb job getting images to the public in a speedy manner, but they are extremely rough, since the team was busy running the spacecraft. However, it is this version that keeps being recycled. Worse, the version on the photojournal is clearly scanned from a printed copy, causing further degradation.



Here is a similar comparison, this time using Proteus (still 1989N-1 on the Planetary Photojournal!).



This discussion started in the thread about Viking crescents but was getting off topic, so I thought I would start a new thread here.

[Stefan]

Here is a similar comparison, this time using Proteus (still 1989N-1 on the Planetary Photojournal!).

Ted, let me first say that I greatly enjoy and admire your work. But I wonder why sometimes it has a bit of a "painted" look (it's almost as if I see "brush strokes"). You can see it clearly in your Borrelly image (saw that one on Emily's blog), and also here on Proteus. Here I actually prefer the original Proteus image on the left, which I think is closer to what I would see with my own eyes. How do you achieve this effect, and are you convinced that all the detail on the right is "real"?

Love your Triton image!

Stefan.

[tedstryk]

Ted, let me first say that I greatly enjoy and admire your work. But I wonder why sometimes it has a bit of a "painted" look (it's almost as if I see "brush strokes"). You can see it clearly in your Borrelly image (saw that one on Emily's blog), and also here on Proteus. Here I actually prefer the original Proteus image on the left, which I think is closer to what I would see with my own eyes. How do you achieve this effect, and are you convinced that all the detail on the right is "real"?

Love your Triton image!

Stefan.

The painted look is from severely underexposed raw data. As a result, high contrast detail is sharp, low contrast detail is not, and is sometimes lost entirely. I am on vacation, but I have a sequence of Proteus that I will post when I get home that shows the Protean features rotating, confirming that they are real. The version on the left hides the problem by blurring the image, but the high contrast fine detail is wiped out in the process.

In the case of Borrelly images (and also in the case of out of focus images), a similar problem is created by desmearing.

ElkGroveDan, I make these from the raw frames on the PDS. The "official" versions came from the Planetary Photojournal.

[chiron]

Hi Ted

I have to agree what Stefan says. I think the normal Images does not contain so much detail. Its easy to say it for Proteus, since there is only one image where this moon is bigger on an image than about 100-150 pixels. Its the frame number C1138920.

A simply Conversion of the PDS Data to the png format without any change is here:
http://www.bernd-leitenberger.de/download/C1138920.png

Proteous is very dark, nearly at the same level as the background, therefore you have much noise on the picture. The planetary photojourna made out of this noise structures and your pictures made much more details out of it - but i think they are not in the original image.

I think you should test your method and validate the results. You can do this by example, if you use the same method with an old viking or mariner 9 image and compare it with an image with better resolution by mex or you work on voyager iamge of the moons of saturn and compare with an cassini image. Without validation, that you really show hidden details and not create only new details from noise, the images are nice to watch but scientific worthless.

Attached thumbnail(s)

 

[Stefan]

Its the frame number C1138920.

Now, I have not worked with Voyager images, and unfortunately lack the time to dive into that. So I simply took that raw GIF image and carefully stretched it:



I also inspected the intensity profile across the moon and noticed that any "detail" on the left side of Proteus is of the same order as the noise in the adjacent dark space. Now I'm sure that careful dark current subtraction and flat fielding will extract more detail from this image, and I would be eager to see the intermediate results.

The detail is very consistent. I wasn't confident of it until I saw the features in the second closest set, which is a multi-frame sequence that I could stack to reduce noise.

The middle, stacked image looks perfectly natural, and I trust all detail to be real. The image on the right, however... from what I see in the raw image I suspect that most of the fine details on the left side of the moon are artifacts. Did you apply some sort of sharpening algorithm or (I hesitate to use the word) deconvolution?

Stefan.

[tedstryk]

If you notice, there aren't interpretable features on the left side except at the very top. There are tantalizing hints, but much less can be seen.

Tasp, it is pretty straightforward. When you process an underexposed image, when you deconvolve an out of focus image, or when you desmear an image, high contrast details are recovered but low contrast ones that are simply not in the data are lost. Hence, high contrast detail is sharp, and low contrast detail is lost. Add to that the fact that in some cases, particularly when you are dealing with an underexposed image, some lower contrast detail can be recovered by binning it 2x2 or even 4x4. However, binning costs spacial resolution. Hence, lower contrast details that are recovered appear amorphous when compared to the high contrast details. That is why the limb looks sharp, as does the area where the illumination angle makes for nice shadows. However, the features look amorphous on the left side because there are no high-contrast features (in other words, no shadows). That is what creates that brush-stroke like effect.

[Stefan]

If you notice, there aren't interpretable features on the left side except at the very top. There are tantalizing hints, but much less can be seen.

Back to the Proteus image... I did some "poor man's dark correction" by subtracting the average of images C1241506 and C1241758 from C1138920 and removed the reseaus. I got the following result:



At left the raw image, in the center the raw minus dark, at right a median filtered version of the center image. All images are stretched (the center and right images by the same amount). If you smooth the right image it would look very much like the official version from the Planetary Photojournal. Ted, I do not see any of the subtle features visible on the left side of your Proteus, and I am now convinced your method introduces artifacts (the "brush strokes") on single, underexposed images like this one. I still prefer the official version of Proteus, although on that one details on the left side are not real either (amplified noise).

One more thing. The images shows very clearly an illuminated feature on the right (dark) side of Proteus:



This must be part of the moon (i.e. the far rim of the big crater), but does not show up in any version I have seen so far! One must be careful when forcing "empty" space to be black.

Stefan.