PDA

View Full Version : Wishlist: Image quality - beyond megapixels


Shadowblade
15th of June 2009 (Mon), 12:25
Let's face it - the megapixel race is running out of steam. Sure, we can produce more and more crowded sensors, with less and less noise, while increasing pixel density, but, as yet, we cannot break the laws of physics. With the latest generation of cameras (5DII, 50D and 500D) we are running into diffraction limitations at larger and larger apertures - f/7.6 in the case of the 50D. Also, we are running into problems with high-resolution sensors outresolving lenses. Cramming more megapixels into a sensor of the same size won't solve anything - all it means is 30 blurry megapixels with the same image quality as a 10-megapixel image and a 30MP file size and noise level. In effect, the only way to increase your image quality further by increasing megapixel count is by increasing the sensor size, with corresponding increases in size, weight and cost, of bodies and lenses.

But this need not mean a slowing of improvement in image quality - raw megapixels are only part of the equation. Hence, my wishlist for Canon's next generation of cameras:

1) How about a 5DIII with 21 true megapixels? Or even 12 true megapixels?

Due to their use of Bayer interpolation, current DSLRs are rather optimistic in their pixel count. A 21MP camera has just 5.25 million 'real' sets of RGB values, while a 12MP camera has just 3 million sets. The rest are interpolated from these sets of RGB values, in a similar vein to extra pixels being interpolated from existing pixels when enlarging an image in Photoshop (but obviously using a different technique). Meanwhile, the 21MP sensor, with only 5.25 million sets of RGB values, suffers the diffraction limits and noise levels associated with cramming 21 megapixels onto a small piece of silicon...

Professional video cameras have not one, but three CCDs - one each for red, green and blue. Each sensor is full-sized, with the same resolution as the final image. After passing through the lens, light is split into three streams, each being captured by a CCD. The three images - one red, one green, one blue - are then combined, in a similar manner to the red, green and blue channels of a digital image in Photoshop. Other sensors (e.g. the Foveon X3) do not require the splitting of incoming light, but which use a three-layered sensor to capture the three coloured images. This results in an image with the same true resolution as the sensor resolution, with no Bayer interpolation required and no corresponding loss of image quality. A 12 megapixel trilayer sensor will produce an image with 12 million 'real' sets of RGB values, not 3 million.

In effect, a 10MP, 3-sensor (or trilayer sensor) image will have larger photosites, a much greater diffraction limit and a smaller file size than a 21MP sensor using Bayer interpolation, while maintaining twice the level of detail of the 21MP sensor.

Which brings me to my next point:

2) Bigger pixels mean less noise and more dynamic range.

This is simple physics. A bigger photosite captures more photons, resulting in a greater difference in light between a full photosite (maximum intensity, or 'white') and a photosite empty apart from random signal noise (minimum intensity, or 'black'). The ratio between maximum and minimum is the dynamic range. Increasing this ratio gives a more vibrant image with greater highlight and shadow detail - often much more important than a nominal increase in megapixel count. A 2-stop increase in dynamic range would result in every photo taken having the dynamic range of a 3-shot HDR (taken at -1, 0 and +1), while a 4-stop increase would result in a similar dynamic range to a 5-shot HDR (taken at -2, -1, 0, +1 and +2). These cover the full dynamic range (or close enough to it) of almost any photo. Similarly, the relatively lower level of random signal noise means lower noise and higher ISO capability.

Naturally, a greater dynamic range means a greater difference between black and white. Since the data is saved in a digital format, this means a greater difference between each level of intensity in the file saved to the CF card. More levels means smaller differences, which mean less banding in the final image - particularly when manipulating the image in DPP, Lightroom, ACR or Photoshop. Fortunately, modern microprocessors are more than up to it. Which brings me to my final point:

3) What about true 16-bit output?

A great deal of photo manipulation this day is done with 16-bit files - photoshop PSD as well as TIFF format. Once saved onto the hard drive in such a format, the camera's 14-bit output will end up taking up the same space as a 16-bit file anyway, with a quarter the number of intensity levels. There is little reason the sensor's data can't be saved in 16-bit format in the first place.

Canon - please produce a full-frame, 12MP DSLR using either three sensors or a trilayer sensor, with true 16-bit output. I'd pay top dollar for it. I don't need thirty million washed-out, noisy, with 22.5 million of them being interpolated anyway. Give me 12 million real, vibrant, clean pixels, with the same level of detail as 48 million Bayer-interpolated pixels (and a greater diffraction limit, less noise, more dynamic range and a smaller file size) any day!

AllenF
30th of June 2009 (Tue), 10:39
How big would that body be to handle those much larger than video sensors and their prisms that will be splitting the light? How much more will this body weigh?

I think 16 bit is coming but comparing Canon to Nikon in the 14 bit offerings Canon is much more refined and ahead IMO. So until Nikon catches up Canon will rest in it's 14 bit laurels.

I must admit I was lost on the true resolution argument mostly because I do not want to take on the research today. :(
If what you say is true, which for now is tough to swallow, then while not at the true native resolution but in some way interpreted it is pretty amazing that there is a truly noticeable improvement with each resolution increase. I say if it is an improvement than what is there to whine about? Sure it could or might be better but they will dole out these improvements a drop at a time especially with fast release cycles just to be able to have something new to offer us each year. Greed and keeping your product fresh in you customers eyes will prevent them from giving us too much too soon. IMO

Shadowblade
30th of June 2009 (Tue), 12:52
How big would that body be to handle those much larger than video sensors and their prisms that will be splitting the light? How much more will this body weigh?

Using a trilayer sensor, there would be no splitting of light, only one sensor and hence on prisms or increased bulk.

I think 16 bit is coming but comparing Canon to Nikon in the 14 bit offerings Canon is much more refined and ahead IMO. So until Nikon catches up Canon will rest in it's 14 bit laurels.

Nikon, hurry up!

I must admit I was lost on the true resolution argument mostly because I do not want to take on the research today. :(
If what you say is true, which for now is tough to swallow, then while not at the true native resolution but in some way interpreted it is pretty amazing that there is a truly noticeable improvement with each resolution increase.

A current 20MP sensor does have 20 million pixels - 5 million red, 10 million green and 5 million blue. What it does not have is 20 million separate red, green and blue values, which is what you would need for a true 20MP image to be displayed on screen. Hence, the software interpolates the data from the existing pixel to give estimated green and blue values for red pixels, red and blue values for green pixels and red and green values for blue pixels.

Also, not all the improvement with each generation of cameras is due to resolution increase - much is due to improved processing of RAW files and improved software interpolation of the red, green and blue values generated by the Bayer matrix. Run the same RAW file through an early RAW converter, then run it again through the latest version of DPP, and you'll likely get a better result from the newer software.

I say if it is an improvement than what is there to whine about? Sure it could or might be better but they will dole out these improvements a drop at a time especially with fast release cycles just to be able to have something new to offer us each year. Greed and keeping your product fresh in you customers eyes will prevent them from giving us too much too soon. IMO

Unfortunate, but probably true...

tkbslc
30th of June 2009 (Tue), 13:26
I think they have given us a lot of improvements beyond megapixels in the last 10 years. Go try out a D30 next to a 50D and lets compare:

-Image detail
-noise above ISO 200
-dynamic range
-AF speed
-AF accuracy
-AF tracking
-burst speed
-processing speed
-customizations
-screen quality




As for your tri-layer sensor theory, is that not what Sigma Foveon does? I agree that the Foveon can look actually more detailed at lower ISO than bayer, but they have to do something about higher iso. It completely falls apart at ISO 800.

Shadowblade
30th of June 2009 (Tue), 14:43
I think they have given us a lot of improvements beyond megapixels in the last 10 years. Go try out a D30 next to a 50D and lets compare:

-Image detail
-noise above ISO 200
-dynamic range
-AF speed
-AF accuracy
-AF tracking
-burst speed
-processing speed
-customizations
-screen quality

I'm not talking about what has been done up till now - I'm talking about what can be done from here onwards, given that further reductions in pixel size are now essentially a game of diminishing returns due to diffraction limits. And, as you have said, noise, dynamic range, AF, etc. have improved greatly since cameras became digital and will undoubtedly continue to do so, as they have yet to run into physical limitations.

What I am also saying, though, is that there are ways to increase the camera's resolving power (as opposed to DR, noise and general functionality) without increasing the sensor's resolution and running into diffraction problems and without increasing the size of the sensor (and, consequently, the size of lenses and other accessories).

As for your tri-layer sensor theory, is that not what Sigma Foveon does? I agree that the Foveon can look actually more detailed at lower ISO than bayer, but they have to do something about higher iso. It completely falls apart at ISO 800.

That's exactly the sensor I am alluding to. Obviously they have a lot to work on, but there is a more potential for development there than in the decades-old Bayer interpolation, which has now run into physical limitations preventing further increases in resolving power without increasing the overall sensor size. The recently-developed panchromatic cells, sensitive to light over the entire visible spectrum, hold similar promise for increasing resolving power without decreasing pixel size.

alessandro2009
12th of July 2009 (Sun), 11:16
For an innovative sensor I hope in Fuji but it quit for the market and for the moment i don't see any productor with a real innovative sensor and with this internation chrisis I think is really improbable investment in this direction.

Canon is also often deaf regarding listening the users unless they decide to cut some heads in the marketing department that makes the good and bad time, I do not think that will come from their innovative proposals.

5DIII
Af (similar nikon)
better dynamic range
less megapixel

Jannie
14th of July 2009 (Tue), 17:44
So, ah, which is the best camera? and why do I think my 10mp MKIII puts out better IQ than my 12mp 5D did? Or maybe it's better if I just like what's coming out of my dream camera and not know all that other stuff lol.

bsmotril
17th of July 2009 (Fri), 08:15
I would like to see something like a 21 Mpixel sensor with a new CPU capable of pixel binning at 2:1 and 3:1 ratios to get better hi ISO noise performance. You would still have the option of very high resolution in good light, but could go down to a 10 or 7 mpixel RAW for bad light hi ISO shooting. I don't think Canon is yet doing all they can in this arena with the RAW options on existing cameras, mainly because it takes a lot of processing power and would put a big dent in frame rates. A newer generation of processors with true binning built into the chip could make this a real possibility, effectively doubling or tripling the size of a photosite on the sensor with respect to signal to noise ratios.

alessandro2009
20th of July 2009 (Mon), 05:31
I would like to see something like a 21 Mpixel sensor with a new CPU capable of pixel binning at 2:1 and 3:1 ratios to get better hi ISO noise performance. You would still have the option of very high resolution in good light, but could go down to a 10 or 7 mpixel RAW for bad light hi ISO shooting. I don't think Canon is yet doing all they can in this arena with the RAW options on existing cameras, mainly because it takes a lot of processing power and would put a big dent in frame rates. A newer generation of processors with true binning built into the chip could make this a real possibility, effectively doubling or tripling the size of a photosite on the sensor with respect to signal to noise ratios.
Sorry but it isn't so simple.
It is impossible have a similar raw.
A new more powerful electronics certainly helps since it allows to use more sophisticated algorithms very useful for example in noise image (but don't only noise image) but if the source (the sensor) isn't good the quality even with a more powerful electronics could not improve much.
If you see image take at 1600 iso you see always a lost of data (even in 800 iso image).
Yes Canon do quite well but that lost is present the same.

Regarding the glass.
Are really capable to use this amount of megapixel?

Shadowblade
22nd of July 2009 (Wed), 04:23
Regarding the glass.
Are really capable to use this amount of megapixel?


Which is one of the reasons I'd be keen on improvements to the Foveon-type design.

You could have a 10MP, 36x24mm sensor, with pixels over twice the size of those on the 5D2 or 1Ds3, providing 10 million red, green and blue values (instead of 10 million green, 5 million red and 5 million blue, interpolated into 20 million RGB values) and thereby giving greater detail in the final image, yet without outresolving lenses or becoming diffraction-limited at common f-stops like current Bayer sensors.

5DIII
Af (similar nikon)
better dynamic range
less megapixel

I'd like to add full weather sealing, bracketing with electronic shutter and simultaneous exposure bracketing (recording several exposures simultaneously by recording sensor data to disk several times, for different shutter speeds) to that list. Megapixels? Don't really care either way... certainly don't need more, but the same would be fine.