H. Irma - blurry images


#1

YooHoo! The tiles seem distorted in the Area 1m96x16y2g and associated tiles. blurry too. Whine, whimper.


TIP - H. Irma #1 Campaign
H. Irma Campaign Now Posted
#2

There seems to be a lot of smeered images posted. Can they be cleaned up?
Example Area 1m96x5y19
TIA
Stuart


#3

Well I don’t know if it’s me just now, or the angle of the images but the buildings seem to be getting smaller/farther away from me now so I guess it’s time I went to bed. I certainly know I’ve over done it now, I’m started to feel sickly:sleepy:


#4

Is it just me or does the Irma tiles seem to be just slightly off register? Seems fuzzy to me even when I go in +2 . And we could definitely use a BOAT tag. This is going to be difficult unless they can sharpen images. How about for you Cagey? Seem blurry to you as well?


#5

If the boat looks like it is damaged (piled atop each other), just use the trash tag. If still in their slips, all nice and neat, just skip those.

I don’t find it blurry. Just overwhelming and difficult to judge what is damage, versus cultural difference, versus poverty… Plus, I’ve never been to islands.


#6

Yes I think there are some blurry areas, but it’s the angle the image was taken that has me struggling at times. Like the satelite has taken these photos upside down and through its legs looking backwards :stuck_out_tongue_closed_eyes: I suppose it’s the best they could get under the circumstances and considering the weather conditions too :confused: It’s better than nothing than nothing at all :wink:


#7

Hi All - thanks for pointing these out. Never a dull moment.

Will you continue to post other smears here? (ie by posting the URL or just the map, ie we have 1m96x5y19).

This may be a result of image processing, or it may be satellite interference, but I’ll need to look into it a bit.


#8

Hi Folks - found out what is causing the smeared (and blurrry/smeared) imagery. These images have a fairly high off-nadir angle, greater than 30% (many of you know what this is, but it’s the satellite looking to the side instead of straight down). When the satellite’s line of sight is off-nadir, it can be hard to match the satellite’s image with a model of the earth’s surface, which is what we do to produce the normal-looking images you see. In this case, there’s nothing we can do to correct it.

Normally, we would not use this image because of this issue. But in disaster response, we want to get going on the issue with whatever we have available, and also collect imagery as quickly as possible, even if conditions aren’t ideal. This is the trade-off we make when we get a relief effort out relatively quickly (in satellite time, that is).

Good (bad) example, posted from @pommystuart above: https://www.tomnod.com/campaign/campaign_2118/map/1m96x5y19


#9

As for the not-smeared but blurry imagery, we noticed this in certain images as well when we were setting up the campaign (we missed the smears somehow). Turns out it is haze, or low clouds, or similar - so again not much we can do there. We’ve already run an algorithm to reduce haze a bit, but we can’t entirely clean it up. Another trade-off in post-storm imagery.

Thanks all for your work and the interesting discussions here!


#10

Thank you Mel for your reply.
Glad you can do something with some of the images.
Keep up the good work. :clap:


#11

Mel_nod thank you. this clarifies things. I understand that the images are stretched by the algorithm due to the Nadir angle. is there a possibility to consider turning off the stretching in such cases the gain from watching the distorted images feels negligible. with unstretched imagery we might be able to comprehend what we see.


#12

From my understanding, anytime DG instructs the satellite to “look backward” over its shoulder, the images will be “stretched” and so, not line up with how we normally envision the earth. The algorithms don’t make the stretching; the high nadir “looking backward” is what does it.

I think of it as putting a coin face up on the floor, walking 25-50 feet away and standing with your back to the coin, and then, look over your shoulder (or between your legs) and see how much of the coin face you can distinguish? Now, breathe air into a balloon and put the coin on top and repeat with looking backward to it. The further away you are changes the nadir, and how “high” or “low” the object is on the “surface” changes how we see it. So they run an algorithm to “smooth out” how we expect to see objects on earth.

You know @Mel_Nod it might help if folks could see one “untreated” strip of a backward looking nadir.

It’s interesting that DG has been working on “cloud thinning”. Please do instruct the algorithm to watch for 1 black cat, yours truly, before thinning the cloud upon which I lie. Please also watch out for my whiskers, as I’ve grown quite fond of my ‘sensitivities’ upon my face. Just don’t want a buzz cut, tell Server Johnny, k?


#13

:laughing::laughing::laughing::laughing:

Now that, I understand perfectly! Thank you cagey


#14

Actually, when I re-read Mel, she said looking sideways. But I figure it’s both similar… but for sideways, do a big :roll_eyes: : first. Then turn your eyes very far to the left or right :eyes: and keep looking that way while you fly off the back of the couch while watching one area on the floor!

Ow-- My head and eyes are hurting just thinking of it.

:earth_americas:

@Mel_Nod Do the satellites also scan right side up, upside down, and… hmmm… while doing a Back Toe Lutz followed by a triple jump? :wink:


#15

:laughing::laughing::laughing::laughing::joy: Oh god cagey my eyes hurt just thinking about trying that :laughing:


#16

Good discussion here! I can’t easily give you an un-processed version of this image - I would need to re-order and process it which is no small feat.

But, if you’re interested - this distortion we’re seeing is introduced in the orthorectification step of the processing, which is one of the first things we do to improve raw satellite imagery both in terms of locational accuracy and to improve visual distortions by aligning to a Digital Elevation Model (DEM) of the area.

So the un-ortho’d image, in this case, would also be quite distorted - just in a different way that this one :wink: Hope that makes some sense!