*UPDATE* 1Ds Mark IV Feature

[unexposure]

I have to add something to my post:

Phase One does it this way (pdf). They practically do an option c) which is: separating green from red and blue into two overlapping "layers" to reduce the gaps in pixel density and as a result can use method a slightly different method a) to bin the pixels. Considering, that the pixel-density of a fullframe-sensor at 45mp would be at least double as high as on a medium-format p45+ sensor, so it might even be possible to use this method whithout rotating the green-layer.

Another Example of how it could be done is shown by fuji's exr technics. Here it's basically another arrangement of the color-array to do 1:2 binning called pixel fusion. The article linked explains it very well.

I guess at the end, the only one who can give a real answer on this issue is canon itself - but even regarding registered patents by canon does not really deliver any clear answer at the current time.

 

[blufox]

One thing I do not understand is that how can Canon move to high MP sensor which require new lenses altogether?

Think for a while
- My 3000$(maybe 6000+$ for some) worth of lenses suddenly will become useless on such a camera instantly.
Woah!... that would be some pretty retarded move by Canon to piss off a lot of people who value glass over body.

I doubt it will be something which requires altogether new lenses.

If not, what did I miss?

I for one, hate MP race. All I need is some good innovation in current 21MP sensor on a 5d MkII.

PS:- I still shoot with a basic 1000D but with a L glass, so I can be safely presumed as a newbie .

 

[kubelik]

to be honest, its not as disturbing as you may think, blufox. as some have pointed out, the pixel density on the current APS-C cameras is already up in the 45 MP equivalent range for a FF sensor.

current L glass isn't "useless" on the 7D, so there's no reason once a 35 MP FF camera comes out that you wouldn't be able to put current L glass on it and get great photos.

it's just a matter of updating and refreshing lenses for the future, in terms of image quality, feature set, and build quality.

from using the new 100 L Macro, and seeing what's soon to come in terms of the revised 300 and 400, I think overall it's a good thing, and nothing to get freaked out over

 

[Osiris30]

Actually noise performance could be improved non-trivially in a binned environment. Basically you are working with 4 times the number of data sources, so the random variance because 1/4 as strong. I'm definitely not using the right terminology right now (you'll have to forgive me, it's been a *very* long day at the office), but I did want to respond with a general direction of what I'm thinking. I'll respond later tonight or tomorrow in more detail if you would like, but long story made very short, you have 4x the signal to 1x the noise in a 4:1 (2x2) bin. That is significant.

I understand that binning increases the signal/noise per pixel, but my point is that the binning doesn't need to be in camera. You can later bin the full resolution raw image in software with the same noise improvement. But perhaps you were referring to some other advantage?

(the S/N will improve 2x for 2x2 binning: the signal increases 4x but the noise only 2x because of the tendency of random noise to "cancel out" when you add it up - that's why adding up 4 noisy pixels only gives twice the noise as a single pixel)

If you bin on chip you cut the readout noise out of the problem. That's the difference between on chip binning (done right) and post production binning. Readout and amp noise can be a non-trivial factor of noise. That's what I was referring to the 4:1 SnR change. On a sensel level the noise is more around 2x, you're correct (I've seen 2.5 used a lot as well). Point in all the is, the earlier in the pixel pipeline you can move the binning, the less 'secondary' or 'system' noise you'll be subjecting those pixels too. Every step (as I'm sure you're aware) exposes the signal to potential degradation.

 

[Osiris30]

Maybe we should have a contest: Who is your favorite unreal photographer?

SPIDERMAN!


ok... something more serious...

a typical bayer pattern would look like this:

so binning for a Bayer sensor would mean, binning pixel-clusters of 2x2 resulting in 4x4 binned-clusters rather than real pixel-binning which would only be possible in similar-to-foveon-type-sensors.

Correct.. which is what I keep calling (rightly or wrongly) 4:1 binning. A 36MP sensor would bin down to 9MP. This is why I don't *not* believe the binning. 9MP in very low light would probably be sufficient detail if it was super clean. Now that the MP count is high enough to support 4:1 I think it's a viable technology.

If you bin the 4 pixel from the same color,you will have a big effect on the sharpness. because the binned pixel will be over lapping the other 2 color. You will end up a blurry picture. Let us face it, if the binning is so easy, All camera manufacturer could have done it. Just remember that 3 MP is good for 8X10. Up to 15 years ago, "professional "digital camera is only 2 MP. So If binning is that easy, they could have given us a 2 stop extra sensitivity and less noise picture at 4.5 MP from a 18 Mp sensor. If it was true. I would take it anytime for the low light situation.

Rocky:

Your logic is flawed on many levels. First of all, some P&S do have various forms of binning and have for years.. so ya, it's doable. Problem is the SLR market place is a different beast. Most magazines, stock photo places and clients will not accept a 4.5MP image in this day and age. It just doesn't have what they need in terms of workable resolution. IMHO 9MP is enough.

Also I'm not sure what you mean overlapping the other two color...

 

[neuroanatomist]

...seeing what's soon to come in terms of the revised 300 and 400, I think overall it's a good thing, and nothing to get freaked out over

Well, to be completely honest, the $11,000 suggested retail price of the new 400mm f/2.8L IS II does sort of freak me out...

 

[Rocky]

Your logic is flawed on many levels. First of all, some P&S do have various forms of binning and have for years.. so ya, it's doable. Problem is the SLR market place is a different beast. Most magazines, stock photo places and clients will not accept a 4.5MP image in this day and age. It just doesn't have what they need in terms of workable resolution. IMHO 9MP is enough.

Also I'm not sure what you mean overlapping the other two color...

Can you name a P&S that have pixel binning? (I am not talking about down sizing after DSP). It should be a big deal for the P&S. I am surprised that Nobody ever advertised it. If you pick 4 pixel with the same color, there are other 2 color in between the 4 pixels. So if you bin the 4 into one. The other color also will be binned. so all the 4 pixel of the same color square are over lapping each other.

 

[neuroanatomist]

Can you name a P&S that have pixel binning?

Several recent PowerShots, cameras from Fuji, Panasonic, etc., all do pixel binning (for the 'low-light' mode). On the other end of the spectrum, Phase One does it, too, in the P65+. One notable difference - all of the sensors with Bayer masks which are capable of binning are CCD, not CMOS.

 

[kubelik]

...seeing what's soon to come in terms of the revised 300 and 400, I think overall it's a good thing, and nothing to get freaked out over

Well, to be completely honest, the $11,000 suggested retail price of the new 400mm f/2.8L IS II does sort of freak me out...

that's true, I can't say I'm ecstatic about the price tag ... but I think there's enough existing 300s and 400s out there to keep people happy for now. maybe canon thinks we're all going to get huge raises over the next few years. or that sports PJ agencies will suddenly start turning huge profits.

maybe they feel that, since Sigma is going "high-end" all of a sudden, canon needs to be "even higher-end" and more expensive/exclusive

 

[epsiloneri]

If you bin on chip you cut the readout noise out of the problem.

If readout noise is indeed significant, then yes, it would be advantageous. But is it? Usually readout noise becomes important for very low S/N in an image, i.e. underexposed parts such as deep shadows, and limits the dynamic range available. I wonder at what exposure level the readout noise becomes comparable to the photon noise.

 

[AJ]

I asked around to find out if software could correct what the lack of an AA filter could bring about. Most notably; moire. The general consensous was it was possible.

I don't think it's possible. Once you get spatial aliasing (moire) the signal is irreparably damaged. Trying to reconstruct the original signal is a non-unique mathematical problem. It's been tried in other signal processing applications where aliasing is much more of a problem. The solution is to keep the AA filter, and to design a filter with cutoffs as sharp as possible right at Nyquist.

For comparison, seismic recordings are always made with an analog AA filter up front. That said, moire is not nearly as extensive a problem as aliasing in seismic recordings.

 

[lightsabre]

My opinion is that while it could be possible to remove moiré artifacts through software algorithms, the process could remove some details as well. My guess is that, to a certain extend, moiré removal has a similar effect to noise removal.

So designers could be faced with two options:

1) Not use an AA filter, leave all the details and moiré but lose some of these details when moiré artifacts are removed by the software
2) Use a light AA filter and lose some details during image capture but get raw images which are mostly free of moiré artifacts

If we look at a similar problem when deciding on Noise Reduction strategy, most would agree that we prefer the camera to leave all the noise in the picture and allow the photographer to decide how much details to sacrifice later during post processing. Software NR allow us to fine tune our personal preferred balance to noise vs details.

Therefore the omission of the AA filter and leaving everything to the software AA algorithm would mean that photographers would be able to decide for themselves the balance between fine details vs moiré. Also, raw files can be reprocessed at a later date with newer and more efficient moiré filters thus squeezing more details out of the image while removing moiré artifacts.

My only question is whether moiré artifacts from a sensor without an AA filter can destroy the details of an image to an extend that cause it to contain less details than a sensor with an optimum AA filter? If this is the case, then my reasoning might not be valid anymore

 

[roger]

Interesting ideas.

But I doubt Canon will do something that revolutionary with 1Ds4. Just doesn't sound like Canon. They always try new technologies on some lower-end cameras instead of taking risks on their flagship models.

 

[neuroanatomist]

I sometimes wonder if magazines set minimum MP limits so that they automatically exclude pictures taken from compact cameras...

Doing a quick google on this subject reveals that some stock agencies will not accept less than 12MP - I'm sure that will jump as soon as all of the pro Nikon gear has more MP.

Try google sometime, yourself...

http://www.google.com/search?source=ig&hl=en&rlz=&q=14-megapixel+camera

Among the search results, which include many makes of P&S cameras, you'll even find a cell phone with a 14 MP camera!

 

[c.d.embrey]

[quote

I sometimes wonder if magazines set minimum MP limits so that they automatically exclude pictures taken from compact cameras...

Doing a quick google on this subject reveals that some stock agencies will not accept less than 12MP - I'm sure that will jump as soon as all of the pro Nikon gear has more MP.

I don't do stock photography, I do advertising. Ads are turned in as 300 dpi PDFs, if the photo in the ad fits that criteria it doesn't matter what it was shot with. A ten megapixel camera will give you in excess of 300 dpi as an 8x10.

From fashion photographer Benjamin Kanarek's blog http://www.benjaminkanarekblog.com/2009/06/18/pixel-peeping-anal-about-number-pixels/

"In fact my last shoot I did with the Pentax K20D was over kill considering the size of the support i.e. about 8×10 inches. In fact I had to reduce the image size to get down to 300 dpi for Pre Press. I would have been quite comfortable with a 8-10 mega pixel DSLR. The only grain I see when published with a 10 mega pixel camera in double page landscape format is the “tram” grain of the printing press of around 133 dpi."

 

[Rocky]

I have to add something to my post:

Phase One does it this way (pdf). They practically do an option c) which is: separating green from red and blue into two overlapping "layers" to reduce the gaps in pixel density and as a result can use method a slightly different method a) to bin the pixels. Considering, that the pixel-density of a fullframe-sensor at 45mp would be at least double as high as on a medium-format p45+ sensor, so it might even be possible to use this method whithout rotating the green-layer.

Another Example of how it could be done is shown by fuji's exr technics. Here it's basically another arrangement of the color-array to do 1:2 binning called pixel fusion. The article linked explains it very well.

I guess at the end, the only one who can give a real answer on this issue is canon itself - but even regarding registered patents by canon does not really deliver any clear answer at the current time.

Thanks. These are the best information that I have ever run into. So far only Fuji and Phase one officially claim the pixel binning. The Panansonic webcam is a totally different animal. the phase one article brings out some very interesting point. when doing a 4 to one pixel binning,the resolution actually decreased to by a factor of 16, due to the over lapping of the binned super pixel. Phase one does not claim the resolution factor. It just says better than 16. From my point of view. Fuji sensor is deigned to be pixel binning. Phase one is kind of a stop-gap. The other point and shoot low light setting may be nothing more than the downsizing of the picture to average out the noise for the pixel peeper to make the picture looks good. Let us face the fact. Pixel binning is a BIG deal. I doubt that Canon or any other camera maker will hind such a BIG selling point from the public.