IPB

Welcome Guest ( Log In | Register )

36 Pages V  « < 2 3 4 5 6 > »   
Reply to this topicStart new topic
MSL Images & Cameras, technical discussions of images, image processing and cameras
mcaplinger
post Aug 28 2012, 04:22 PM
Post #46


Senior Member
****

Group: Members
Posts: 1825
Joined: 13-September 05
Member No.: 497



QUOTE (Airbag @ Aug 28 2012, 09:17 AM) *
But photon shot noise is less of an issue with larger number of photons...

shot noise = sqrt(signal), so shot noise is higher at higher signal levels. The square root encoding is coarser at higher levels, finer at lower levels.

You could have a philosophical debate about what this means, but that's the way our cameras work.


--------------------
Disclaimer: This post is based on public information only. Any opinions are my own.
Go to the top of the page
 
+Quote Post
Airbag
post Aug 28 2012, 06:30 PM
Post #47


Member
***

Group: Members
Posts: 408
Joined: 3-August 05
Member No.: 453



I was looking at shot noise from a S/N point of view but I get the hint and will skip any debates - thanks for the clarifications everybody!

Airbag
Go to the top of the page
 
+Quote Post
jmknapp
post Aug 28 2012, 07:47 PM
Post #48


Senior Member
****

Group: Members
Posts: 1465
Joined: 9-February 04
From: Columbus OH USA
Member No.: 13



QUOTE (mcaplinger @ Aug 28 2012, 08:48 AM) *
The MAHLI paper I've referred to several times before is an excellent source of this kind of information.


http://rd.springer.com/article/10.1007/s11...4/fulltext.html

From another paper....

QUOTE
MAHLI shares common electronics, detector, and
software designs with the MSL Mars Descent Imager
(MARDI) and the two Mast Cameras (Mastcam).


Ahh, good to know. Wonder what the (optimum) SNR is.


--------------------
Go to the top of the page
 
+Quote Post
mcaplinger
post Aug 28 2012, 08:25 PM
Post #49


Senior Member
****

Group: Members
Posts: 1825
Joined: 13-September 05
Member No.: 497



QUOTE (jmknapp @ Aug 28 2012, 12:47 PM) *
Wonder what the (optimum) SNR is.

I often tell people who are fond of such numeric metrics as SNR, MTF, ENOB, etc, that no matter how optimal those numbers make your camera sound, anybody can tell a good image from a bad image. I think the MMM images hold up pretty well.

That said, you could work out the best possible SNR we could get from the Truesense datasheet. You can't ever do better than sqrt(fullwell) for a single measurement due to shot noise and these sensors have a fullwell of 20K-40K electrons so 140:1 to 200:1 is as good as it could be.


--------------------
Disclaimer: This post is based on public information only. Any opinions are my own.
Go to the top of the page
 
+Quote Post
jmknapp
post Aug 29 2012, 12:32 AM
Post #50


Senior Member
****

Group: Members
Posts: 1465
Joined: 9-February 04
From: Columbus OH USA
Member No.: 13



QUOTE (mcaplinger @ Aug 28 2012, 04:25 PM) *
these sensors have a fullwell of 20K-40K electrons so 140:1 to 200:1 is as good as it could be.


200:1 is about 46 dB SNR which according to the method on ctecphotonics gives effective number of bits (ENOB) of (46-2)/6 = 7.3 bits. That could be improved if desired by stacking?

The images are very impressive to see--just wondering what the limits are when the images are considered as abstract data, perhaps to be analyzed by machines teasing out the last bit of information.


--------------------
Go to the top of the page
 
+Quote Post
Airbag
post Aug 29 2012, 06:44 PM
Post #51


Member
***

Group: Members
Posts: 408
Joined: 3-August 05
Member No.: 453



That explains it all Joe - thanks! This made me wonder if the public "raw" images we see are uncompanded (and rescaled into 8 bits) again or not?

As an experiment, I used some full size ML and MR color images of the sundial, which has gray rings of 20%, 40% and 60% reflectivity, and measured the corresponding grayscale pixel values using "sample merged" data point of appropriate radii in the Gimp. Making various assumptions about JPEG accuracy, lighting and dust etc. (i.e. ignoring them!), the resulting data shows that the JPEGs we see appear to be quite linear in response to the different grayscale rings. This suggest that the images we see has been decompanded - unless either my measurements are flawed (for instance, the extrapolation of the trend does not go through the origin), or the CCD is not really linear in response, and the MSL side companding has now made it appear linear?

Attached Image


Comments welcome, as if I could stop them anyway :-)

Airbag
Go to the top of the page
 
+Quote Post
jmknapp
post Aug 29 2012, 07:01 PM
Post #52


Senior Member
****

Group: Members
Posts: 1465
Joined: 9-February 04
From: Columbus OH USA
Member No.: 13



QUOTE (Airbag @ Aug 29 2012, 02:44 PM) *
... the JPEGs we see appear to be quite linear in response to the different grayscale rings.


Interesting use of a test pattern. I gather the square law companding from 12-bit to 8-bit would be something like this:

Attached Image


The image you used ranged from about 100-160--not sure if you'd notice the nonlinearity per the graph above. I.e., does it only matter if there's a lot of dynamic range in the image, otherwise it can be pretty much fixed up with linear levels adjustments?


--------------------
Go to the top of the page
 
+Quote Post
Airbag
post Aug 29 2012, 07:29 PM
Post #53


Member
***

Group: Members
Posts: 408
Joined: 3-August 05
Member No.: 453



Joe, I think you may have the wrong "law" - it should be square root, not square, right? Doesn't change your plot though, and maybe that is just a semantics thing.

You may well be correct that the dynamic range of my samples (from 5 different images) is not sufficient to show whether the resulting data is truly linear or not. Plus, even a little bit of dust would cause more forward scattering and that could be why my data points don't go through the origin. Arguably. I could have included some (presumed) "full white" and "full black" (the gnomon?) points too, but I have no data for actual reflectivity values for anything other than the gray rings.

Airbag
Go to the top of the page
 
+Quote Post
fredk
post Aug 29 2012, 07:37 PM
Post #54


Senior Member
****

Group: Members
Posts: 3893
Joined: 17-January 05
Member No.: 152



This is interesting. A suggestion: look at the calibrated pds sundial images from the earliest MER sols - presumably we know that those 10bit files have linear response?
Go to the top of the page
 
+Quote Post
mcaplinger
post Aug 29 2012, 07:48 PM
Post #55


Senior Member
****

Group: Members
Posts: 1825
Joined: 13-September 05
Member No.: 497



QUOTE (Airbag @ Aug 29 2012, 11:44 AM) *
This made me wonder if the public "raw" images we see are uncompanded (and rescaled into 8 bits) again or not?

As far as I know they aren't decompanded or stretched. Of course you can't tell if they were commanded with sqroot or some flavor of linear (it's an option) so I caution everyone against trying to do photometry.

You could look at ftp://pdsimage2.wr.usgs.gov/cdroms/Mars_R...nt/marcisis.txt to see the companding table used for MRO/MARCI. Why would they be different?


--------------------
Disclaimer: This post is based on public information only. Any opinions are my own.
Go to the top of the page
 
+Quote Post
RoverDriver
post Aug 29 2012, 07:54 PM
Post #56


Member
***

Group: Admin
Posts: 967
Joined: 29-September 06
From: Pasadena, CA - USA
Member No.: 1200



I don't know, but is it possible these images use a gamma value other than 1?

Paolo


--------------------
Disclaimer: all opinions, ideas and information included here are my own,and should not be intended to represent opinion or policy of my employer.
Go to the top of the page
 
+Quote Post
ugordan
post Aug 29 2012, 09:02 PM
Post #57


Senior Member
****

Group: Members
Posts: 3631
Joined: 1-October 05
From: Croatia
Member No.: 523



QUOTE (RoverDriver @ Aug 29 2012, 09:54 PM) *
I don't know, but is it possible these images use a gamma value other than 1?

If the MSSS cameras use sqrt encoding, that's pretty close to an inverse of the 2.2 display gamma so while the encoded DNs don't scale linearly with scene brightness, the apparent brightness scaling on the screen should follow the actual scene pretty closely.

The navcams and hazcams on the other hand look to me like they use a linear 12 to 8 conversion (even though sqrt should be an option for them as well), because they look rather contrast-enhanced.


--------------------
Go to the top of the page
 
+Quote Post
jmknapp
post Sep 8 2012, 12:12 AM
Post #58


Senior Member
****

Group: Members
Posts: 1465
Joined: 9-February 04
From: Columbus OH USA
Member No.: 13



Anyone know why the "full frame" raw MASTCAM images are sometimes 1648x1200, usually 1536x1152, rarely 1600x1200? I believe the specified image size for the camera is 1648x1200. Are the smaller sizes scaled or cropped?


--------------------
Go to the top of the page
 
+Quote Post
Deimos
post Sep 8 2012, 12:49 AM
Post #59


Martian Photographer
***

Group: Members
Posts: 324
Joined: 3-March 05
Member No.: 183



QUOTE (jmknapp @ Sep 8 2012, 01:12 AM) *
Anyone know why the "full frame" raw MASTCAM images are sometimes 1648x1200, usually 1536x1152, rarely 1600x1200? I believe the specified image size for the camera is 1648x1200. Are the smaller sizes scaled or cropped?

True full frame is as you say. 48 columns are dark or null (see the MAHLI paper that has been linked several times). The corners are quite vignetted, so for some purposes 1600x1200 may not be what is desired. For other purposes, it is handy to have multiples of 64 for the size; multiples of 8 are required. Those come from JPEG compression of both the thumbnail and the image. I'd speculate that once we're well past CAP2, you'll see NxM * 128 most often, and 1648x1200 more rarely, and others quite rarely. But you never know.
Go to the top of the page
 
+Quote Post
jmknapp
post Sep 8 2012, 01:37 AM
Post #60


Senior Member
****

Group: Members
Posts: 1465
Joined: 9-February 04
From: Columbus OH USA
Member No.: 13



QUOTE (Deimos @ Sep 7 2012, 08:49 PM) *
But you never know.


Thanks... I've been working out a method to unwarp the images based on the CAHVOR camera specification for MASTCAM. Hard to say how to apply it to the cropped images though. I guess scale 1536x1152 up to 1600x1200 and then add the 48 dark columns back in.

It's easier with the NAVCAM images because they at least come in at 1024x1024 as specified. I'm wondering if stitches would work a little better using such rectified images. Here's a before and after of a NAVCAM shot--it's a real subtle correction:

Attached Image


--------------------
Go to the top of the page
 
+Quote Post

36 Pages V  « < 2 3 4 5 6 > » 
Reply to this topicStart new topic

 



RSS Lo-Fi Version Time is now: 16th January 2019 - 02:44 PM
RULES AND GUIDELINES
Please read the Forum Rules and Guidelines before posting.

IMAGE COPYRIGHT
Images posted on UnmannedSpaceflight.com may be copyrighted. Do not reproduce without permission. Read here for further information on space images and copyright.

OPINIONS AND MODERATION
Opinions expressed on UnmannedSpaceflight.com are those of the individual posters and do not necessarily reflect the opinions of UnmannedSpaceflight.com or The Planetary Society. The all-volunteer UnmannedSpaceflight.com moderation team is wholly independent of The Planetary Society. The Planetary Society has no influence over decisions made by the UnmannedSpaceflight.com moderators.
SUPPORT THE FORUM
Unmannedspaceflight.com is a project of the Planetary Society and is funded by donations from visitors and members. Help keep this forum up and running by contributing here.