Welcome to our community

Be a part of something great, join today!

Log in
  • Hey all, just changed over the backend after 15 years I figured time to give it a bit of an update, its probably gonna be a bit weird for most of you and i am sure there is a few bugs to work out but it should kinda work the same as before... hopefully :)

General Demosaicing Questions

Eric Lange said:
charts that seem to “transend” or go substantially beyond the Nyquist limit …
Click to expand...

This reminds me the modem speed theory based on the same principle.
You can’t exceed modem speed on the transmission line above the half the speed of bandwidth allocated to the transmission line, all based on the Nyquist theory.

http://en.wikipedia.org/wiki/Modem

Well Harry Nyquist is dead (1889-1976) and coincidently, right after his death, first modem exceeding the speed of the frequency of the line was invented. So much for the Nyquist.

He forgot that you can send information by manipulating not only the amplitude but also the phase and temporal characteristics of the signal.
Well at present we are sending data rate exceeding 50 times sampling rate and this is without error correction. With the sophisticated error correction we go even some 5 times more or so.

Just imagine what could be achieved in the sensor resolution domain if we use temporal information to reassemble the image or the photon phase information hitting the photosite or just to use logical pixel placement error correction. (dont't mix it up with interpolation)

Well this is the future of photo sensor technology I guess but Nyquist is long dead.
 
Andrew M. said:
This reminds me the modem speed theory based on the same principle.
You can’t exceed modem speed on the transmission line above the half the speed of bandwidth allocated to the transmission line, all based on the Nyquist theory.

http://en.wikipedia.org/wiki/Modem

Well Harry Nyquist is dead (1889-1976) and coincidently, right after his death, first modem exceeding the speed of the frequency of the line was invented. So much for the Nyquist.

He forgot that you can send information by manipulating not only the amplitude but also the phase and temporal characteristics of the signal.
Well at present we are sending data rate exceeding 50 times sampling rate and this is without error correction. With the sophisticated error correction we go some even 5 times more or so.

Just imagine what could be achieved in the sensor resolution domain if we use temporal information to reassemble the image or the photon phase information hitting the photosite.

Well this is the future of photo sensor technology I guess but Nyquist is long dead.
Click to expand...

That’s really sweet, I like that, that will get one’s imagination going (at least in the frequency domain) and a few others perhaps.

I like it.
 
Trouble is that the aliasing problem is really an optical issue, not an electronic or software issue -- you've got a grid of photosites causing an interference pattern with fine lines in the subject. So the solution tends to be optical as well -- low-pass filtering -- though as pixel resolutions get higher and photosites smaller and less "coarse" the interactions happen on a smaller scale.

But it's hard to say that all it takes is some clever software design to get around interference patterns when the phenomenon is optical, it's not happening post-sensor in the signal chain or later.
 
Eric Lange said:
This is what I had in mind, Perhaps you can clarify a little as to how the envelope was pushed, (without giving too much away), i.e. say nice things about what you and RED team are doing to push things to new limits..
Click to expand...

I did note when I started this thread I wasn't going to go into algorithms. I'm talking generally for educational purposes, not specifically the nuances and algorithms that RED employs as they're rather good and rather proprietary.

Graeme
 
David Mullen ASC said:
Trouble is that the aliasing problem is really an optical issue, not an electronic or software issue -- you've got a grid of photosites causing an interference pattern with fine lines in the subject. So the solution tends to be optical as well -- low-pass filtering -- though as pixel resolutions get higher and photosites smaller and less "coarse" the interactions happen on a smaller scale.

But it's hard to say that all it takes is some clever software design to get around interference patterns when the phenomenon is optical, it's not happening post-sensor in the signal chain or later.
Click to expand...

Yes, an optical + sampling problem, that occurs all before the photons are converted to pixels, and that's the best place to deal with it.

Once aliasing get strong enough to be noticed in an image, it's strong enough that it's removal will most likely cause even more degradation to the image than leaving it there. That's the curse of aliasing....

Also, if you're going into temporal codec, it's good to note that they really hate aliases as aliases move backwards to the motion of the object and all the motion analysis on the codec gets confused and throws a wobbly. That's what the BBC are so cautious about aliasing in cameras as they "eat" transmission bandwidth for no image benefit.

Graeme
 
I think it's quite telling that the Bayer pattern was patented in 1976 and has yet to be surpassed. Sure, plenty have tried:

Sony introduced an "RGBE" filter in 2003, which was essentially a Bayer pattern with half of the green filters changed to "emerald." It was used in one camera in 2004 then was never heard of again.

Kodak filed for some "RGBW" patents in 2007, but I'm unaware of any production sensors being built with the design. It is somewhat promising for cameras with high pixel counts but small photosites, like point and shoots.

I own a Foveon-sensored camera myself (which I am trying to sell, incidentally) and while the colour resolution can be great, you really only see it on subjects with strong red, blue or purple, while the cross-contamination of colour channels really hurts colour accuracy, and the noise at high ISOs has been quite atrocious. I've seen huge leaps forward in the technology of processing the raw files, but I doubt it will ever be enough.

I have to say, the Bayer pattern is one of those exceedingly simple ideas that is really stunningly brilliant, and one of its greatest beauties is an irreducible simplicity that can not be improved by making it more complex.
 
Graeme, we have static aliasing and dynamic aliasing.
I really hate when fine stripes pattern jacket is slowly panned on HD camera and whole texture is pulsating. This is most visible and ugly looking problem.
In contrary the static aliasing, especially in the movie is not that visible.

If you compare value of the pixel in slowly moving object in every odd frame and you do the same in every even frame and you remove any harmonics, you should be able to cancel the dynamic aliasing completely.

Sure if nothing is moving in the frame than no temporal info on harmonic frequencies is there but also no ugly pulsating of aliasing.

Just thinking...
 
Stephen Strangways said:
I think it's quite telling that the Bayer pattern was patented in 1976 and has yet to be surpassed.
Click to expand...

Yes, it's rather enduring for a good reason - it's a very efficient use of pixels to create a great looking image. Much of it's power lies in it's subtlety, and a poor demosaic algorithm is enough to bias people against the images, but a good algorithm can look superb.

I find it rather amusing that one arm of Sony used to trash Bayer, while another arm of Sony make some very successful DSLRs which use Bayer patterns! Similarly the Canon / Panavision talk trashes Bayer, but Canon make a rather successful line of DSLRs all using Bayer patterns. You'd think they'd talk to each other to get their story straight.

Now we have Sony touting their Q67 as the "solution", but upon our analysis, Q67 is just a Bayer pattern, and their "solution" boils down to "double the number of pixels".

It's a funny universe we live in....

Graeme
 
Andrew, when moire moves, it moves more than the object that caused the moire to move in the first place. This aspect was used in some micrometers I think as the moire amplified the apparent movement and made the device more accurate. I think I first saw that in an Open University programme on UK TV back in the 70s (you can guess my age back then) but I could be mistaken - it might have been the 80s, but either way, it's funy how you remember these things....

I remember as a kid being driven down the motorway near our house, and seeing the over-head signs that are held in place over the road by a mesh structure, which produced the most beautiful moire! It took me a while to find out what the effect was called and how it was caused, but I've been interested in moire and aliasing since I was a child.

Graeme
 
Graeme Nattress said:
Andrew, when moire moves, it moves more than the object that caused the moire to move in the first place. This aspect was used in some micrometers I think as the moire amplified the apparent movement and made the device more accurate. I think I first saw that in an Open University programme on UK TV back in the 70s (you can guess my age back then) but I could be mistaken - it might have been the 80s, but either way, it's funy how you remember these things....

I remember as a kid being driven down the motorway near our house, and seeing the over-head signs that are held in place over the road by a mesh structure, which produced the most beautiful moire! It took me a while to find out what the effect was called and how it was caused, but I've been interested in moire and aliasing since I was a child.

Graeme
Click to expand...

This explains a lot.

I just turned 40 this past year. When I was a kid I used to watch the Open University Science and Math shows too! That was the only option you had, as believe it or not (and most American’s will find hard to believe), back then we only had three channels, and during the daytime two channels would be switched off or only broadcast as a Test Card. As a bored kid glued to the telly, you could either listen to really bad music on the test card, or watch some guy with a beard and gravy stains on his velvet jacket, and national health specs, give a lecture on non-Euclidian geometry. (I like the one’s on Einstein riding motor cycles painted blue and red for dopler shift). I think I learnt a lot more then, from BBC2 than school.

Rather ironically the test card on the other channel does in fact offer a hell of a lot of information about aliasing artifacts that still haunt me today…and the whole question of De-bayer.

Maybe that’s why I felt the need to attack the CRT (as a kid) with a magnet to see what would happen as well as other metal objects.. Hey, no-one will notice those weird permanently stuck green and pink “clouds” on the TV, will they?

British telly in the 70's has a lot to answer for…

www.youtube.com/watch?v=Ncmg3bxtTB0

unbelievable, the 1978 BBC test card on you tube!!!, it kicks in about 50 seconds in, (here’s a hint, the clown never gets any) Hey but the resolution targets are VERY informative.

Check this out;
www.youtube.com/watch?v=PEFd4tCWtvk

Graeme, are you sure this is not a relative of yours, the strong correlation between the appreciation of good booze and complex mathematics,

No wonder we are warped…

Fry and Laurie captured the look and spirit rather well I think,

www.youtube.com/watch?v=2un9rO2ZF4g&feature=related




[BTW, I don’t know about the micrometer, but interference patterns are used for accurate laser depth ranging, as well as other dimensional/wave base measuring systems.]
 
I've used interference patterns created by cross tuning delays in voice lift audio systems to reduce feedback modes between fixed pattern ceiling mics and speakers in rooms to increase gain.
Audio moire.:biggrin:
 
This Panavision presentation was originally created in 2008.
http://media.panavision.com/ScreeningRoom/Screening_Room/Demystifying_Part7_480p.html

In the last part (7) they stated that Bayer patter is basically 4:2:2
Also they are stating that if Bayer is claimed 4:4:4 than it has to be extrapolated. Like there would be no small oversampling possible.
In contrary the RGB stripe patter is 4:4:4.

Well allot has changed since 2008 I guess.

I remember the time when real engineer was using only slider for calculation, (no calculators were invented yet) We used to say 2X2 is 3.9999 presume 4 :-)

Without small oversampling and with good algorithm Bayer patter is kind of the same 3.9999 :-)
 
Andrew M. said:
Well allot has changed since 2008 I guess.
Click to expand...

Nothing has changed. Their analysis was wrong then, and it is wrong now. Dig up a link to the old thread....

Graeme
 
Cail Young said:
Are you thinking of Vernier scales?
Click to expand...

From what I remember, and this was a long time back..... It was a moire based optical distance measurer with two pieces of transparent plastic with an array of black markings on them such that they would moire. One of the pieces of transparent plastic was attached to the movement, the other static. I think there was a light source and some means of sensing the light on the other side of the moire device. That's about all I can remember.

Graeme
 
One thing I can’t get from Sony presentation of the future road to 4K.
http://pro.sony.com/bbsc/video/related-recently_added/video-cinealta_hdworld2010_4k_roadmap

He mentioned that delivery will be MXF SStP 3.6Gbits/sec for 2K 10bits (in future 12 bits)

So this is 1.62TBytes/hour possible recording data amount at this speed.

Also he mentioned that future SR memory for recording of above will be 1T and will hold 1hour of recording. He mentioned maximum 50fps at 17MP 12 bits each frame size so this is 4.59TBytes/hour one channel 12 bits RAW.

So he can use 5:1 compression here, so far all clicks.

However it means that it can`t be 12 bit RGB 3 channel raw coming out of the camera unless they will compress data farther 3 times for total of 15:1 compression ratio just to conform to 3.6Gbits per sec speed and 1TByte per hour storage commitment.

As he mentioned, they will have many challenges to overcome switching from CCDs to CMOS.
He mentioned, floor noise, global shutter, demosaic patterns.
He can add this one to the equation as well. 15:1 compression and still keep the “image quality”
I guess even Sony concluded that CMOS is way to go, not without the challenges though.



Andrew
 
From a casual read thru of the redcode patent, the other guys may be a severely limited by Red's IP as to how they deal with compressing raw data from a planar sensor of any kind.
 
David Rasberry said:
From a casual read thru of the redcode patent, the other guys may be a severely limited by Red's IP as to how they deal with compressing raw data from a planar sensor of any kind.
Click to expand...

That was my thoughts as well...

OTOH... RED hasn't patented wavelet compression, have they?

Or maybe that's exactly what was described?
 
At least as far as I understand it, the magic lies in how interpolation of raw pixel data is parsed to eliminate redundant data from the file. So Sony ET AL may be locked into demosaic and encode to an RGB pixel format prior to compression. Which is pretty much what they have always done anyway.
 
The fanniest thing is that after glorifying CCDs for so long they concluded that they have to switch to CMOS now for 4K and just to justify past bashing of CMOS they show how many chalanges lay ehead of them with CMOS.

Welcome to the club!

Both CMOS club and 4K club.
This is the honour to have them here.
 
You must log in or register to reply here.
Back
Top