Who Stole My Pixels?

[WarrenMars]

No doubt many readers know this already but I was truly SHOCKED to discover that each supposed pixel in my camera contains only the data for one colour! This means that the true resolution of my camera is only a QUARTER of its advertised value! My 10MP SLR is really only 2.5MP!

I have been aware of interpolation and the demosaicing step for years, but I had always assumed that this was to make small realignments to centre the 4 elements of the Bayer mosaic within the pixel. It never occurred to me that it was being used to invent colour data for three quarters of the image!!! It was only when I was stepping through the main data structure at runtime in Dave Coffin's DCRAW code and wondering why only one quarter of it was filled in that the penny finally dropped. Shocked I tell you!  

Ken Rockwell is right. The manufacturers are LYING! Interpolated pixels are not real pixels. Made up data is not real data. We have all been conned. Sure, the interpolation used these days is very good! In areas without sudden transitions the invented detail is accurate, but in areas of fine detail: Ouch! It isn't!  

There are laws about misrepresenting the specs of one's products. A minor inflation of one's resolution might be acceptable, but 400%? Surely that's illegal, not to mention immoral.

[dwdallam]

Interesting. I can't wait to see where this thread goes. I would really like more information.

[Plekto]

Actually, it's ~.66 per dimension since it's in a triangular arrangement.  With good interpolation software, it might edge up to about .7x in each dimension, but that's the limit of a Bayer pattern.  It's a quick and mostly effective kludge to turn analog into digital results, much like how you see screens used in offset printing(just take a magnifying glass to any printed magazine)

Your 10MP DSLR is something like 3800X2600 pixels, give or take?  So that's really about 3800X 0.7 X 2800X 0.7.  Roughly 2650X1950(5.2 actual full-color locations).   35mm film is typically scanned in a DLab to roughly 6.7MP - or roughly 12-14MP depending on which Digital camera you're comparing it to.

It's a known problem, but it's not nearly as bad as you make it out to be.

[joofa]

This means that the true resolution of my camera is only a QUARTER of its advertised value! My 10MP SLR is really only 2.5MP!

Since, those 2.5 MP grids are phase-offset, that means that each of them have some unique information and they may be processed together, synergetically, to produce an output that is quite higher than the individual 2.5 MP. It is not unsurprising that one may be able to get 6-8 MP back, which is of course, much higher than 2.5 MP.

BTW, in a different, but somewhat related domain, fewer numbers than expected does not necessarily mean loss in reconstruction. Think about this: If you know a sine wave's frequency, amplitude, and phase, just 3 numbers, you can obtain infinite values of that wave. So an otherwise infinite storage requires only 3 numbers to reconstruct that.

[David Sutton]

To view, or not to view- that is the question:
Whether 'tis safer for the mind to suffer
The mounting horror of outrageous posting,
Or to takes arms against a sea of opinions,
And, by touching the “shut down” switch, end them.

No, I'll bite. Warren, life is mostly lies, beginning with your eyesight. Rather poor information is taken from the eye and constructed into reality in your head, and the brain Photoshops out the dodgy bits. It's life. Get over it.
On the other hand, while I'm not too worried about accurate colour and rather want predictable colour, I too can't wait to see where this goes.

[Bart_van_der_Wolf]

No doubt many readers know this already but I was truly SHOCKED to discover that each supposed pixel in my camera contains only the data for one colour!

This (the approx. fact, not the discovery) is not exactly true, and half (or quarter  ) truths are dangerous.

Each sensel holds the result of an approximate 1/3rd of the relevant visual spectrum, but (usually) after an Optical Low Pass Filter (OLPF, AKA AA-filter) has done its work. For chromatic information that's not too much of a chore to reconstruct, reconstucting luminance information, that will require a bit more cleverness.

This means that the true resolution of my camera is only a QUARTER of its advertised value! My 10MP SLR is really only 2.5MP!

Wrong. I suggest trying to measuring the actual resolution after a good CFA demosaicing. Sure R/G/B resolution individually will not achieve its theoretical maximum for the pixel density, but it will meet (or even exceed!) the sampling density.

I have been aware of interpolation and the demosaicing step for years, but I had always assumed that this was to make small realignments to centre the 4 elements of the Bayer mosaic within the pixel.

That would be a very poor way to demosaic images. The Bayer CFA is much more clever in design, and implementation.

It never occurred to me that it was being used to invent colour data for three quarters of the image!!!

Wrong again, it doesn't "invent" color data, it estimates the missing data from clues in its immediate surroundings (and it does so in a very convincing manner, especially for chrominance).

It was only when I was stepping through the main data structure at runtime in Dave Coffin's DCRAW code and wondering why only one quarter of it was filled in that the penny finally dropped. Shocked I tell you!  

Too bad it required studying Dave's code. It's mostly common knowledge, at least around these quarters ...

Cheers,
Bart

[tokengirl]

This means that the true resolution of my camera is only a QUARTER of its advertised value! My 10MP SLR is really only 2.5MP!

What?

Seriously.  Did your pictures lose 3/4 of their appeal after you found this out?

[elf]

No doubt many readers know this already but I was truly SHOCKED to discover that each supposed pixel in my camera contains only the data for one colour! This means that the true resolution of my camera is only a QUARTER of its advertised value! My 10MP SLR is really only 2.5MP!

I have been aware of interpolation and the demosaicing step for years, but I had always assumed that this was to make small realignments to centre the 4 elements of the Bayer mosaic within the pixel. It never occurred to me that it was being used to invent colour data for three quarters of the image!!! It was only when I was stepping through the main data structure at runtime in Dave Coffin's DCRAW code and wondering why only one quarter of it was filled in that the penny finally dropped. Shocked I tell you!  

Ken Rockwell is right. The manufacturers are LYING! Interpolated pixels are not real pixels. Made up data is not real data. We have all been conned. Sure, the interpolation used these days is very good! In areas without sudden transitions the invented detail is accurate, but in areas of fine detail: Ouch! It isn't!  

There are laws about misrepresenting the specs of one's products. A minor inflation of one's resolution might be acceptable, but 400%? Surely that's illegal, not to mention immoral.

Have you tried counting the pixels in an image to really see how many there are?  I think it's more likely that manufacturer's are telling the truth as they see it and Ken Rockwell doesn't have a clue

[Schewe]

Ken Rockwell is right. The manufacturers are LYING! Interpolated pixels are not real pixels. Made up data is not real data. We have all been conned. Sure, the interpolation used these days is very good! In areas without sudden transitions the invented detail is accurate, but in areas of fine detail: Ouch! It isn't!  

Wow bud...guess you don't know too much about old Ken do ya...if you did, you wouldn't be pointing him out as a raw of truth (unless you were actually being sarcastic).

Really, you just discovered what a Bayer array is? Boy, you live a sheltered life...and ya know, you really SHOULD NOT believe everything you read on the internet...of wait, I think you just figured that out recently huh?

 :~)

[Wayne Fox]

Ken Rockwell is right.

Now there's your problem ... he'll say absolutely anything to get hits (that's how he makes his money) , and I'm not sure he has a clue about the actual technology itself.

[Gigapixel]

Actually one can estimate the lower bound for the perceived resolution by taking the number of green pixels, which in a Bayer pattern accounts for 50 percent of the total number of pixels. To the human eye green represents the major part of luminance and therefore sharpness information. In combination with interpolation from the red and blue pixels the achieved "real" resolution lies roughly between 60 and 70 percent of the total pixel count.

[Bart_van_der_Wolf]

Actually one can estimate the lower bound for the perceived resolution by taking the number of green pixels, which in a Bayer pattern accounts for 50 percent of the total number of pixels. To the human eye green represents the major part of luminance and therefore sharpness information. In combination with interpolation from the red and blue pixels the achieved "real" resolution lies roughly between 60 and 70 percent of the total pixel count.

That depends on the subject, and on the demosaicing technology used, but it is usually better than people think, or want us to believe. I've tried to determine the real loss of resolution, and it can be as little as 6.4% for luminance only. Here is a simple experiment I documented 6 years ago. This was before I got hold of real (deconvolution) sharpening routines instead of simple edge contrast enhancements like USM.

Cheers,
Bart

[Roger Calixto]

What?

Seriously.  Did your pictures lose 3/4 of their appeal after you found this out?

All prints 75% off =) Dontcha love a post that has nothing to say? (I'm referring to myself here)


On the other hand, just take it for face value. Your prints haven't gotten any smaller and your photos will still look exactly the same.

[PierreVandevenne]

No doubt many readers know this already but I was truly SHOCKED to discover that each supposed pixel in my camera contains only the data for one colour! This means that the true resolution of my camera is only a QUARTER of its advertised value! My 10MP SLR is really only 2.5MP!
Ken Rockwell is right. The manufacturers are LYING! Interpolated pixels are not real pixels. Made up data is not real data. We have all been conned. Sure, the interpolation used these days is very good! In areas without sudden transitions the invented detail is accurate, but in areas of fine detail: Ouch! It isn't!  

Technically, the manufacturers are right. The pixels are there, and there's just a filter array in front of them. Many CCDs are available in both versions, with and without a CFA. If the issue is really of interest to you, rather take a top-down approach I would suggest a bottom-up approach. The astronomy CCD books give a solid foundation, but it isn't mandatory to buy them. There are good resources on the web. The "fishing for photon" serie by Craig Stark is a great start http://www.cloudynights.com/category.php?category_id=181

Another great resource can be found here

http://www.olympusmicro.com/primer/java/index.html

with very nice animated tutorials.

When you are familiar with that aspect of things, you move up to "fancier" sensors such as CMOS, add CFAs, look into the methods used to minimize read noise, etc...

One thing to keep in mind is that since the digital imaging field is so hot scientifically, commercially and has a lot of military applications, tens of thousands of great minds are at work in their own subfields. Unfortunately Ken Rockwell isn't in that category. He is a fairly decent entertainer who makes strong statements aimed at an unsophisticated audience. In some cases, he is just plain wrong. In other cases, he states the obvious and makes a mountain out of it. In a way, he's the web equivalent of the UK tabloid, fishing for audience at any cost.  Keep in mind that tabloids can be right at times, as some pro golfer learned.

[stamper]

I think this might be the antidote to his state of agitation?

http://www.foveon.com/article.php?a=67

Funnily enough I haven't read anything about an image taken with this sensor being superior to one of the "normal" ones?

[dwdallam]

One thing to keep in mind is that since the digital imaging field is so hot scientifically, commercially and has a lot of military applications, tens of thousands of great minds are at work in their own subfields. Unfortunately Ken Rockwell isn't in that category. He is a fairly decent entertainer who makes strong statements aimed at an unsophisticated audience. In some cases, he is just plain wrong. In other cases, he states the obvious and makes a mountain out of it. In a way, he's the web equivalent of the UK tabloid, fishing for audience at any cost.  Keep in mind that tabloids can be right at times, as some pro golfer learned.

No, he's the photography equivalent of Rush Limbaugh, without the hostility.

[AlexB2010]

Bayer pattern is the industry standard for color photography, counting the individual pixels aren’t cheat. They are using this since the beginning. Some comparison of individual bayer pixels can’t be aligned with resolution in lpmm. There is some dark ages misinformation on comparing film and digital. On digital realm that’s the standard. Non bayer sensors, foveons or B&W are the exceptions.
Best,
Alex

[NikoJorj]

I think this might be the antidote to his state of agitation?

Ouch! Second time warp in less than 20 posts! That DOES hurt!

Better turn it on the side to get a transmogrifier than playing with the time machine... Oh! No! A duplicator! This thread is duplicated now!
(with due respect)

[dwdallam]

I think this might be the antidote to his state of agitation?

http://www.foveon.com/article.php?a=67

Funnily enough I haven't read anything about an image taken with this sensor being superior to one of the "normal" ones?

Well, pretty amazing that Foveon. So what's the scoop on this image sensor? Good, bad, better?

[Jeremy Payne]

No doubt many readers know this already but I was truly SHOCKED to discover that each supposed pixel in my camera contains only the data for one colour!

[!--quoteo(post=0:date=:name=WarrenMars)--][div class=\'quotetop\']QUOTE (WarrenMars)[div class=\'quotemain\'][!--quotec--]I am one of the few photographers who really understand the nitty gritty of how these cameras work[/quote]

Which is it, Warren?