How many pixals is enough? revisited

[jim monroe]

About a week ago someone on this forum brought up the question of how many pixals was enough to get a quality shot? I believe they were mostly interested in wedding pictures and thought the 10D's approximately 6 Meg were enough. Others had differing opinions.
First my experience with human nature tells me for many there will never be enough, sort of the same as asking how fast does my computer need to run? One seems to always wants a faster computer even when it seems to me to be completely unneeded.
But back to how many pixals. Is my following analysis basically correct. I want to photograph, amongst many other things, tow boats and barges moving up and down the Ohio River. A single barge being pushed by a towboat is approximately 300 ft. long. So the 10D having about a 2000 by 3000 pixal display means I get 10 pixals per foot of view. Lets kick that up to 12 pixals per foot of view. So 1 pixal per inch.
Like many people on this forum I look for real detail for many reasons. So suppose there is a gasoline tank on the boat with the word gasoline on the tank. I want to be able to read that in my photo. The lettering suppose is 3 or 4 inches tall. There is no way that I can see with 1 pixal/inch that I can ever read this in a photo. The e in gasoline would need at the very minimum 5 pixals vertically to be distinquishable but if the lettering is 4 inches tall we have only 4. Really to have any chance of reading the e I would think I would need about 10 pixals vertically over the space taken up by the e, This means for this level of resolution, disregarding any blurring by lense, focus, etc. I need about 2 to 3 pixals/inch and hence overall would need at least 24 million pixals and realistically maybe up to 54 million pixals
Again is my locic correct?
To carry this one step further my guess is that there will be rapid progress over the next several years regarding producing higher pixal count cameras at increasingly low prices. At some point the resolution powers of the lense must come into play, disregarding the ever present focusing problems.
Anyone have any ideas on how many pixals one needs before the lense resolution becomes the main problem?
Seems to me whatever that is then progress will grind to a halt although it may take a ridiculous number of pixals for this to matter.

[design crusader]

Not to be a bother, but it is not pixals.

The word is pixel. I just figured if you were going to be crazy about them...

Created from two words -- picture and element.

[BearSummer]

Hi Jim,

Ok so let me see if I understand this right, boat is 300 feet long, text is 3 inches high, you want to be able to print it and read the text.... ok so lets say the text once its printed is 1/8 of an inch high, which with reasonable eyesight should be able to be read as if reading from a book.... that means the print will be 12.5 feet long and your nose is about 12 inches from it... not the best viewing position. Using the rule of thumb that you should be the diagonal distance of the print away from it to view it correctly, you should be 15 feet away. 1/8" letters at 15', now thats good eyesight.

Regarding cameras out performing lenses. We are quite close to that already with the 1DS, there are several articles about "requiring L glass as normal glass underperforms". Without serious "wonder" lenses (L+) I dont see much point going beyond 15Mp as the glass will be the limiting factor IMHO.

Best Regards

BearSummer

[lightandlife]

Jim, your logic is sound, but the numbers may confuse the majority.

I have tried to follow your logic and simplified it for others here. Currently, 10D's sensor can record 6 Mega pixels. It is 2000 x 3000 = 6 million pixels.

Suppose one used a smallest aperture so that 99% of the picuture is sharply focused. At this stage, the lens can see clearly everything or "pass through" the image to the sensor accurately. As in satelite photos, one is somtimes interested in details, and focus on 1% of the image area, i.e., 1% crop, and still see everying inside this area as if one had a telephoto lense focused in that area.

This means 100 x 6million or 600 million pixels. I think Canon's L glass would still see things clearly, but may impose limitations beyond this stage.

Alternatively, one can explain this in terms of enlargement. Typical prints are 4" x 5", and many enthusiasts like to enlarge them to poster size (which is about 100 times 4" x 5"). It is not the lens, but the grains of film that would make the prints grainy.

So, I would say 100 million to 1 billion pixels would put an upper bound for poster size enlargement and scientific purposes.

Within the next 10 years, we will reach that. If we double pixels every two years, then it will be (6, 12, 24, 48, 100 million) in 10 years. We were raising pixels much faster during the past couple of years.

[Rayz]

It's not just the number pf pixels but the quality of the pixels that counts. By quality I mean efficiency; the ability to record all the photons that pass through the lens.

Let's say each photodetector (pixel when interpolated) is 100% efficient, you'd still be limited by the resolution of the lens and every lens, no matter how good, is ultimately limited by diffraction. Astronomers use a formula, derived from a chap called Rayleigh, to calculate the maximum resolution of a lens at a specific aperture. If you want to include all wavelengths of light, the maximum resolution of a lens at f8 is 1000/8 = 125 lp/mm. At F2 the maximum resolution would be 100/2 = 500 lp/mm, but no lens is that good. Other aberrations take over at such wide apertures.

It takes a minimum of 2 pixels to capture one line pair, in theory. In practice, more. So, if we're talking about a theoretically perfect situation, you'll need 250 pixels/mm to capture everything an excellent lens can deliver at F8.

The 10D sensor size is 22.7mmx15.1mm. At 250 pixels/mm we get 5675x3775 = 21.4MP.

Unfortunately, it wouldn't be possible to make such small photodetectors efficient enough to capture those 125 lp/mm because the contrast of the lines is so low - probably about 10% of their original contrast. Those who understand MTF understand the problems.

[Derek Smith]

Yes of course we need more pixels. The 10D is the first affordable back to give a level of detail that just starts to be professionally usable, yet even so, it still takes considerable effort to get workable images from crops or parts of whole images. Yet the advantages of the digital route over scanned film are so great that we are prepared to use work-arounds like thinking ahead of the probable use and taking full frame shots of areas that are going to require particular clarity in the final image.

Today's PC processing power can easily crunch through 200MB image files, and the new DVD RAM disks can store over 4GB, so the logistics of handling much larger files is not a limiting factor, so yes, when higher resolution backs roll out at these prices, we will be wanting them.

But for me, the first and most important 'want' in a camera of this price, is a full sized 35mm chip (preferably square). The jump to a full sized chip from the 22.7mm chip in the 10D would ramp up the resolution to over 9 Mpixels - very handy!!

However, the real reason for wanting the 35mm chip, is because I would like to get full value out of my '35mm' glass. All of our lenses are designed to give good linear image reproduction over a circle of 42mm diameter (corner to corner across a 35mm film). Sitting a little 22.7mm chip in the middle of this expensively produced image is wasting nearly a third of the lenses capability and its the third that cost the most to achieve.

So here is the challenge to Cannon. Working with today's existing CCD resolution sensors, develop a circular sensor 42mm in diameter together with a file storage format for circular images. Without any further increases in sensor technology, this would yeild a whomping great 25Mpixels of image from my precious '35mm' glass.

In photography, 'Glass is God' - it is our real investment. Backs will come and go while good quality glass will be used over and over again. So please Cannon - lets stop wasting all those precious edges. Stop thinking 'rectangular landscape' and start thinking truly digital. With a round sensor, every way is up and the only way that matters is how the image is processed before output.

Go on - Dare to be innovators!! I want to use my glass - I want a round image.

Derek Smith

[lightandlife]

Derek Smith wrote:

In photography, 'Glass is God' - it is our real investment. Backs will come and go while good quality glass will be used over and over again. So please Cannon - lets stop wasting all those precious edges. Stop thinking 'rectangular landscape' and start thinking truly digital. With a round sensor, every way is up and the only way that matters is how the image is processed before output.

Go on - Dare to be innovators!! I want to use my glass - I want a round image.

Derek Smith

A brilliant idea, Derek.
Since the lens aperture is round or a polygon, it makes sense to make a round sensor.

Camera designers chose a rectangualr sensor, just to replace films, which by necessity had to be rectangular. This is probably the reason why, regardless of lenses used, four corners will be softer than the center of any side.

Designers should not be free from the traditions of film, and boldy design a circular sensor for round lenses.

A square sensor for round lenses just don't make sense. Photoshop will have to revamp the program for round cropping also.

[Rayz]

Nice idea Dereck, but would a circular sensor really give us any advantage over a square sensor, apart from the ability to print circular photos with a greater number of pixels. It seems a luxury that few would be prepared to pay for. Imagine the high cost of paper for a start - stamping out these non-standard circles that would presumably need to be stuck to a rectangular backing sheet for printers to accept them.

An amusing scenario, but totally impractical!

There's also the issue of peripheral light fall-off. Even expensive lenses often show some light fall-off in the corners. Cheaper lenses, especially at full aperture, show light fall off in the corners and edges. A circular image would show very noticeable light fall-off around the entire circumference, not to mention resolution fall-off.

[freelans]

If I had a choice, I would opt for square lenses instead of a round sensor if that were the issue.
A round sensor would result in a uniform degradation in the edges rather than any enhancement.
Neat idea though. :)

[Hatem Eldoronki]

freelans wrote:
If I had a choice, I would opt for square lenses instead of a round sensor if that were the issue.
A round sensor would result in a uniform degradation in the edges rather than any enhancement.
Neat idea though. :)
You got my vote! I don't plan on buying a round monitor to view my round pix!
I actually never thought about how a circular lens give us a rectangular picture :) Funny stuff..

[AJSJones]

Derek Smith wrote:

However, the real reason for wanting the 35mm chip, is because I would like to get full value out of my '35mm' glass. All of our lenses are designed to give good linear image reproduction over a circle of 42mm diameter (corner to corner across a 35mm film). Sitting a little 22.7mm chip in the middle of this expensively produced image is wasting nearly a third of the lenses capability and its the third that cost the most to achieve.

Derek Smith

Derek, Have you looked at the MTF charts that Canon provides for their lenses? The part that is being "wasted" is ALWAYS the worst part of the lens's performance (and it is indeed sometimes terrible), even though you may be right that it was expensive to get it even that good! The extra price you have to pay to get "back" your wide-angle capability is offset by two things - you get to use the sweet part of ALL your lenses and you get the "free" telephoto benefits from the 1.6 format factor.

Not only that, you can still use all the lenses. Perhaps the introduction of the Canon EF mount was before your purchase of your "expensive" glass, but the owners of the predecessor mount (the FD) couldn't even put their lenses on the "new and improved" EOS EF camera system.

So life ain't so bad !

Andy

[lightandlife]

freelans wrote:
A round sensor would result in a uniform degradation in the edges rather than any enhancement.
Neat idea though. :)

4 corners are softer than the middle of any side. A round sensor would eliminate these 4 soft corners and expand the mid sections of each side.

Man, there is always a trade-off. A round sensor eliminates the soft corners and expand the sharper image area. So with a round sensor, you get a larger sharper area, say 20% more than a square sensor.

First, note that lenses are round. Suppose one could measure sharpness by the distance between the center and a point on the image area, zero being the sharpest and 1 being the barely acceptable sharpness. A sensor of 2" x 3" will give you only a sharp circle with radius 1. All other points outside the area would be blurred. What a waste of digital resources? With such images one would crop a square inside the circle. It is like a square orange with thick skin, and the juicy part is a really small globe inside. Some Japanese scientists tried to market square (actually cubic) watermelons unsuccessfully, although it was easy to pile them up, like films.

I think less than half the shots taken are actually printed. If the pics are for Internet or e-mailing, round pics may become handy.

Trying to get a square image from a round lens is fundamentally, and philosophically wrong. But inventors did it because of square films. Now we are not bound by square films. Digital should free camera designers.

[Rayz]

lightandlife wrote:

Trying to get a square image from a round lens is fundamentally, and philosophically wrong. But inventors did it because of square films. Now we are not bound by square films. Digital should free camera designers.


And trying to make a square lens is much more difficult. If it wasn't, we'd have them. We live in a predominantly square world of man-made objects; rectangular houses, walls, paintings, TV screens, printers, loudspeakers, books etc, etc, etc. Most things are squarish and if they're not, like aeroplanes and lenses, there's a good practical reason for it.

[lightandlife]

There is a good reason for not producing square lenses.
Cost!

Also there were many good reasons for producing square camera (meaning box). Not any more for digital sensors. We can still house them in squares or cubes.

[RWChimself]

Having recently researched this issue with specific reference to my desire to persuade my personal financial enabler to fund me a 1Ds. I have formed the opinion that the latest Kodak medium format back at 16Mp would well and truly satisfy many fellow photo-thesbians BUT of course the image "size" of this device is bigger than the 1Ds SO one assumes that a print made to a good exhibition size would look "better" from a 16Mp DCR Back than1Ds which in turn might suggest that a 1Ds with a 24Mp sensor i.e. 1/3 more pixells than the Kodak joy machine could well match it in every way except one - namely selling price, which one would most earnestly hope was much lower.

Let's have more pixells and have done with it, oh and add in a 24 bit dynamic range as well, but not for the next 10 years because after that time, and being a penioner I will be too old to care.