Pixels

[ChristianG]

OK, so at least one person persuaded me to do this.

I'm going to make an attempt, given that some of you appear unprepared to read my friend Charles Maurer's article, to try to deliver a (relatively) short explanation of what constitutes a pixel (picture element). For this purpose I'm going to make some assumptions:

* I am here referring only to what are generally described as "Point & Shoot" (P&S) cameras, I am not referring to dSLRs even though the same principles also hold for them - except if they use the Foveon sensor, as in the Sigma SD14, when all bets are off but that's another subject entirely.
* All P&Ss use CCD (Charge Coupled Device) or CMOS (Complementary Metal-Oxide Semiconductor) sensors with an area of either 7.18mm x 5.32mm or 5.76mm x 4.29mm. The only exception to this is still vapourware so I'll ignore it. For purpose of this little exercise CCD and CMOS are interchangeable.

Do you know what RGB stands for? Red, Green and Blue, that is what your monitor employs to create the illusion of colour and so does your P&S. Of course RGB are very specific shades of those colours because between them they create all those other colours - including white and black (which are actually not colours at all). In a sensor (as described above) four cells (which the camera manufacturers, all of them, quite incorrectly and very misleadingly, count as a pixel) are employed to create a pixel and these cells are respectively sensitive to R, G, B and G. That collection of cells properly forms a pixel in a mosaic pattern so you can all divide by four to arrive at a proper pixel count for your own camera. This method is employed using the Bayer (after it's inventor) principle.

There is a problem here: because each pixel (as described above) creates a unique shade of colour the one immediately next to it may, often does, seem rather different to its neighbour so all these cameras use a blurring filter to overcome this problem. IOW, what you see is NOT what you get. The blurring algorithm is actually pretty good but it is nevertheless a blurring algorithm.

Now let's address the issue of printing and here there are also some difficulties because there is absolutely no point in trying to print something which the eye can't see - there is lots that the unaided eye can't see. For this purpose I am going to use A4 (297 x 210mm) as the benchmark. Note that A4 is more than double the size of a conventional, large, photograph.

Incidentally, and as an aside, when the Germans introduced this series of paper sizes to the world (in 1922) they were extremely clever about it because all the sizes are consistent with the proportion of 1 to the Square Root of 2. This is the only rectangle which allows you to reduce and enlarge an image (of whatever type) in direct proportion. No other rectangle does this and I am, quite simply, astounded that with a new era in photography the manufacturers of these sensors, all of them, chose to ignore this concept. They should, all of them, be hung, drawn and quartered for that.

The amount of processing power required to resolve an A4 image to print (trust me, I'm a retired printer) is 1,600 by 2,400 pixels (proper, real pixels - as described above) or 3.8 (real) megapixels, a far cry from what the camera manufacturers would have you believe and most of you wouldn't want to go beyond that size - if you did you'd be a contender for a decent dSLR which will almost inevitably have a bigger sensor than those described above. So even if your camera outputs more pixels than the above your eye will quite certainly not be able to discern any higher quality than that - except perhaps under a loupe, but who other than printers examines pictures under a loupe? Those extra pixels are only useful to you if you are editing in Photoshop or whatever which is why RAW is a desirable feature on cameras, especially for u/w use.

Hello? Just 3.8 megapixels will properly resolve into an A4 photograph? Just 1.9 megapixels will resolve into an A5 image (half A4 - 210 x 148.5 mm) which is rather closer to conventional photograph size but still a little larger? Do the camera manufacturers really want you to know that?

Remember what I said about sensor sizes? Here we come to noise, which is when a set of cells is so small as to be wellnigh useless:

Simply put it is just plain stupid to cram more cells into a finite space.

A (poor) analogy would be a double keyboard piano. Yup, lots more notes but will there ever be anyone that can properly control such a beast? Much more importantly:

the smaller the cell the more resistant it is to the acceptance of both wavelength (colour) and light, thus noise is created.

So, for example if you are looking at the purchase of a camera and the two you lust after have the same (theoretical) pixel count then, all other things being equal, you should choose the camera with the 7.18mm x 5.32mm rather than the one with the 5.76mm x 4.29mm sensor simply because the cells will be larger on the former camera. Now that's a really important consideration which should always be kept in mind.

More is not necessarily better than less. Less = larger cells = better sensitivity to light and wavelengths = larger pixels often of better colour saturation = well, less is actually more in many of the instances of P&S cameras - even sometimes dSLRs.

Caveat: This is a very simplified way to look at all this. To see the whole picture, read, and re-read my friend Charles Maurer's article:

TidBITS : Sense & Sensors in Digital Photography

which goes way beyond the scope of what I have written here.

If you go to that extent, you may also wish to read the other articles he has here.

[Cussy]

Christian,
Very interesting post, I agree 100% with you.

To back up your comments many have noted here on YD the reported low light characteristics of the Fuji F50 versus the F31 - this is simply because they have attempted to cram too many pixels onto a chip, IMO. So many that each cell is no longer independent; if one cell has a high value it will bleed into the surrounding cells. This is counteracted by filtering, which reduces the sensitivity to light.

Many of us have said time and time again about P&S cameras "it is not the number of mega-pixels, it is the quality of the glass in the lens." I would also add to that the features that are also useful underwater (RAW, manual WB etc).

I think this was also shown recently in the dSLR world, when the latest top of the range Canon has 20MP+, yet Nikon decided to invest their development in features (mainly in catch up to Canon). Many eyebrows were raised.

Rob

PS Has anyone done comparisons with the Sigma? This sounds like there are 4 times as many pixels as most other cameras.

[ChristianG]

Originally Posted by Cussy

PS Has anyone done comparisons with the Sigma? This sounds like there are 4 times as many pixels as most other cameras.

Chuckle.

I took delivery of a Sigma SD14 Tuesday a week ago and I take delivery of one Saeed Rashid, he of photo manipulation expertise at your local dive shows, tomorrow. Saeed shoots with a Canon EOS400 which is roughly, pricewise, equivalent to my shiny new Sigma (sorry, sadly no housing available) so we might try some like for like comparisons.

My friend Charles Maurer describes all of this much better than I ever could in these articles:

TidBITS : How to Buy a Digital Camera
TidBITS : Colour & Computers
TidBITS : Editing Photographs for the Perfectionist

and if the above leaves you hungry for more:

TidBITS Advanced Search

Sigma is as guilty as every other manufacturer in that it describes photocells as pixels (if only to "compete" with everyone else), except that in the case of the Foveon sensor there is no extra "G" photocell thus you need to divide by three rather than four to obtain the proper pixel count. This is, BTW, no bad thing given that the Foveon photocells are arrayed very differently to those of every other CCD/CMOS sensor:

Foveon - What's a Pixel
Foveon - Why X3 is Better
Foveon - Image Comparison

[David P]

Originally Posted by Cussy

many have noted here on YD the reported low light characteristics of the Fuji F50 versus the F31 - this is simply because they have attempted to cram too many pixels onto a chip, IMO

Indeed - I'm very annoyed with the needless growth in pixels. The F40/F45 is much more attractive than the F47 and certainly the F50 as a result. The reviewer's sad comments in the conclusions at Fujifilm FinePix F50fd Review: 1. Introduction: Digital Photography Review - effectively, "we hoped for better from Fuji, but were disappointed" - are telling. The F31 is a great compromise in terms of low light performance vs. quality, and seemingly a high point that's unlikely to be reached again.

Me - I hope pixels alone will never be a buying factor. My Canon 30D has quite enough... It's back to "what the camera does with them".

Nice article, Christian, thanks. Useful sanity.

[Cussy]

Originally Posted by ChristianG

Sigma is as guilty as every other manufacturer in that it describes photocells as pixels (if only to "compete" with everyone else), except that in the case of the Foveon sensor there is no extra "G" photocell thus you need to divide by three rather than four to obtain the proper pixel count.

That means, although I tried carefully to find this out on the Sigma site the other week, I fell for the hype too (the website is too Flashy and not enough content). I read it as 10MP (or whatever it has) with an R, G, and B at each pixel.
Rob

[Cussy]

Originally Posted by Cussy

PS Has anyone done comparisons with the Sigma?

Just found this Foveon X3 Sensor Claims Put to the Test - - PopPhotoJuly 2007 . As ever, for any benefit there is a slight drawback in other areas. In the UK where my images are dominated by green's and not far-off b&w detail, the Foveon might not be beneficial.

I still think though that this is an interesting technology. But I think manufacturers will just continue to pile up the pixels and hope for improvements based on redundancy on the amount of required pixels.
Rob

[ChristianG]

Originally Posted by Cussy

Just found this Foveon X3 Sensor Claims Put to the Test - - PopPhotoJuly 2007 . As ever, for any benefit there is a slight drawback in other areas. In the UK where my images are dominated by green's and not far-off b&w detail, the Foveon might not be beneficial.

I still think though that this is an interesting technology. But I think manufacturers will just continue to pile up the pixels and hope for improvements based on redundancy on the amount of required pixels.

Rob,

Have a look here:

Sigma - News

Now I know that this is publicised by Sigma themselves however they may even be putting their necks on the line. Whose to say what they know about the conclusions reached by Shutterbug?

I, for one, will be watching that with keen interest.

[Jeremy H]

Interesting post ChristianG.

Can I just clarify something...

resolve an A4 image to print (trust me, I'm a retired printer) is 1,600 by 2,400 pixels

There doesn't seem to be mention of an output resolution in your post. Are we assuming that all the images under discusion here are being output at screen (72dpi) or say up to 150dpi. Say for inkjet or web work as opposed to magazine etc repro? (At say 300dpi I estimate we'd get an A5 image (not allowing for selective cropping etc) - or have I missed something?)

[Cussy]

But a dot does not always equal a pixel. In monochrome terms a dot is either black or white; dithering makes grey. So can takes quite a few dots to make a single pixel.

As so many magazine cover shots have been taken with either a D100 or D70, it does show that a modern dSLR (or compact for that matter) has far more pixels than it needs. In my living room I have a 7MP cropped photo from an Olympus 7070 blown up to an A1 photo and it still looks fine. Admittedly if you are right up next to it it's not perfect.

I've been sending photos off for printing since my first digital in '99. That was a sub mega-pixel camera (Olympus C1000) and you could not tell it was a digital at 6 x 4 or 7 x 5.
Rob

[Jeremy H]

Hi Rob, yep you're right. A printing dot doesn't equal a pixel.

For a consumer (prosumer?) market then I agree you don't need to worry about more pixels. I've no doubt that a 6mp SLR can produce a cracking large format output, particulary if it's printed using an inkjet type printer - and I also agree that just squeezing more elements into a ccd without making it bigger isn't going to lead to a corresponding increase in quality.

I'm simply suggesting that as when discussing resolution the eventual output size (and output intent) has to considered. The choice of camera depends on the need and the quality required. Surely though, for certain usages a 6mp sensor won't always cut the mustard...

A good example I came across recently was the BBC Countryfile calender. People didn't worry about print etc - they just took a 'good' picture with the camera they had. Howver, there were several shots in it that just did not hold up when commercially printed. They looked great from across the room but close up some were pretty bad. The best solution would have been to print the pics at different sizes rather than to 'res them up' to the same size throughout. (Or insist on a 5-6mp minimum for entry! )

Also the nature of shot has to be taken into account - you'll probably get away with a 6mp action shot or a soft landscape as a mag front cover (and a bit of photoshop unsharp masking etc) but perhaps not a studio shot with lots of fine detail. (The commercial photographers I know tend to shoot a lot higher than 6mp - for example - an A2 poster size at 260ppi is about 70(ish)mb in RGB.) You really do need a big sensor for that kind of thing.