Flutterby™! : Canon 1Ds 2002-09-12 01:46:56+00

Canon 1Ds

2002-09-12 01:46:56+00 by Dan Lyke 26 comments

Aaron passed along as "confirmed as legit" this announcement that includes the Canon EOS-1Ds

The EOS-1Ds features the first ever full frame 35mm CMOS sensor with 11.1 million effective pixels. The EOS-1Ds capture astounding detail & colour, almost doubling the resolution ordinarily considered state of the art for a digital SLR camera in the world today.

It appears that the selling price will be about $5500.

Wow. I look forward to buying one for $200 in another five years or so. Film is fun, but film looks increasingly dead.

Mars, the only reason film isn't completely dead for me is that I can buy a 2CR5 lithium disposable battery in any drug store, and it gives me lots of rolls of film, and a spare is good in my camera bag for 5 years. The rechargeables in my digital cameras have a storage life of about 5 days after I've charged them, and tend to go dead at the worst possible moments.

Oh yeah, other reason to like film: I can push 5-7 frames/second sustained for 36 frames through my EOS-3 body. I've been waiting a lot recently to transfer data around the D-30 I've been using.

I keep hoping to hear about the Foveon guys breaking out into the mainstream. Still waiting...

Standard sensors have discrete color sensors, (1 red:2 green:1 blue) spread out across their surface, meaning that at max resolution you never get true color sampling at any pixel. Yes, this means a 4 megapixel camera only samples red 1 million times. The software tricks used to smear out the color info among surrounding pixels have visible artifacts that give any full-resolution image from one of these sensors an unavoidable "digital camera" signature, most obvious as a purple fringe in certain super-high contrast light/dark transitions, which is not due to chromatic aberration in the lens. Foveon has created chips that stack the red, green and blue sensors at each pixel location on the chip, giving true full color info for each pixel, avoiding the processing artifacts, and reducing power requirements by avoiding the calculations used to smear out color info among the pixels.

I read somewhere (no cite) that 35 mm is good for up to 14 megapixels (and of course thats full color achieved by stacking different chemical layers for different colors of light, a direct analog to Foveon's digital approach. That means that just to match full-color pixel counts, current style digital sensors will have to get to 56 megapixels before they match 35 mm. That still looks to be a good ways off, without even considering other factors like range of sensitivity, image noise, etc. (And keep in that film, while far ahead of current digital tech in sensitivity ranges, is still nothing compared to what the human eye can see.)

Consumer print films have resolution in the 50-60 lines/mm range. Velvia, one of my favorite slide films, runs 80 limes/mm. Nominally, that means that if you're shooting Velvia (ASA 50), you can expect to get 160 pixels/mm, or 3840 by 5760 for a 35mm (24x36mm) frame, 22 megapixels.

It's even possible to get higher resolution than that, but that generally requires specialized B&W film. However...

Getting over 60 lines/mm also means you've got to be using prime lenses or very expensive zoom lenses. Even more restrictive than that, you've got to be using those lenses within their sweet spots, for zooms that usually limits the focal lengths you can be using, and the aperture ranges are likely to be limited to the F4 or F5.6 region.

But the big deal is that all the sweet glass in the world won't save you from camera shake. If you're not using a tripod, and if you're not the sort of person who thinks about mirror lock-up and the like, then you're not getting much more than 50 lines/mm. (Strongly recommend John B. Williams Image Clarity if you want to delve into the details.)

Even if you've got all that resolution, the first thing the viewer will be aware of is grain. Grain is intrusive at far lower frequencies than film resolution. The reason you can see an image on the wall, or even in a magazine, and say "that was shot on 35mm film" is grain. Long before you get to the 20"x14" enlargement resolution where the resolution would start to make a difference, the sky starts to look mottled, you notice skin texture which isn't pores, and so forth. Grain, not resolution, is the main reason why people shoot medium format.

Digital has artifacts, but film grain, which I'm very aware of, is not one of them. I've always been happy with 3k by 2k scans of my film, so if I could get that in glorious RGB I'd consider film dead. I'm perfectly aware that my eyes read green far out of proportion to red and blue, so I'm okay with Bayer sampling (that puts two green pixels in the sensor for each red and blue pixel), and I really only find that my eyes are sensitive to the results for certain diagonals, so I'd even be happy with a Canon D60. But the 1Ds should have enough oversampling in the color that it'd get everything I've ever hoped to get out of 35mm film.

All of that is the long way of saying "I could believe you could get 22 megapixels of resolution out of 35mm color slide film. I don't believe many people are, and I think that considerations other than resolution come into play at about 6 megapixels."

When dealing with the resultant product (a developed negative), what's the distinction between resolution and grain?

Grain is variation or texture in color that should be constant, resolution is how fine a line a region can hold. If you had a .005 wide (on the negative) line, that line would be visible, but the line wouldn't necessarily be constant color.

True dat. Would it be correct to equate resolution and grain in B&W photography?

technical discussions aside. I bought a Sony 707 (5mpix) on the way to BurningMan...

Note: These are approx 2mb images.. don't click on these unless you are a fast link. But for a yahoo like me to get good pics like this, I think that the age of digital camera's are here to stay. My camera gets a usable 120 minutes on a battery, and stores hundreds of pics... OK.. It ain't cheap yet, but it will be..

Dan: That 22 megapixel number is actually pretty accurate. Please don't forget the megapixel rating on most cameras is for SINGLE COLOR pixels; your 6 million pixel estimate for the resolution of a 35mm frame is about right if those pixels are 24-bit color. Digital cameras interpolate their data to get that 6 million pixel shot out of their 8-bit sensors, and suffer in resolution and sharpness. A film scanner has no such limitation. A camera with 6 million 24-bit sensor spots would be considered an 18 megapixel camera; a 22 megapixel camera would have the equivalent of 7.3 million 24-bit pixels.

What I really like about Canon's new camera is its full-frame sensor. I'm very fond of wide-angle shots, and the 2/3 sensor size on my current digicam is disappointing.

And.. as my camera has gotten better, I am realizing how important, lighting, a tripod, and practicing 'click' more like shooting a gun becomes important. ie: breath control. holding camera steady, SQUEEZING lightly the shutter button. Biggest complaint. although the camera does manual focus, the digital viewfinder/screen is not high enough res to really do manual focus correctly. I took over a thousand shots at the Burn and the trip. A lot of them are trash. At some point.. the tools surpasse the casual user. It's time for me to work on technique.

Resolution and image quality aside, some of the most important areas where digital cameras still trail their 35mm counterparts are:

Problems (1,2) can be solved with better (faster) internal memory management and buffering. Problem (3) won't be solved until full-frame image sensors become affordable.

frame to frame.. my 707 will do bursts of 3 frames per squeeze with a selectable time in between shots.. and you can select an auto exposure variance between steps as well.

again, the 707 has a two step shutter, squeeze a little, it will auto-whatever and then the second step seems to be pretty darn instant. It is frustrating when you just want a quick squeeze shot though. it pauses while it figures things out unless it's in manual mode.

re: FOV (field of view) do you want MORE of the fisheye effect or less? a minor fish eye wide angle is one of the things I want to add..

Your image-to-image bursts are pretty darn impressive. The biggest problem that I have with my little Olympus D-510 is missing the spontaneous shooting opportunities. There is discernable latency between the time I press the shutter release and when the exposure actually occurs.

FOV Example: 20mm lens with FOV multiplier of 1.6 = 32mm lens. This is a pain for me since I enjoy the juxtaposition of near and far in many of my compositions.

In contrast to Norman Koren, Robert Monaghan takes the stance that film is better than digital in many ways.

It's funny that Monaghan uses National Geographic as his benchmark, even with the smaller sized pages I've found grain intrusive on those images, and given the limitations of offset printing would far prefer that they trade-off some resolution for grain. Of course he also makes a comment about "Leica lenses", if his attitude is similar to most of the folks I've run into who carry a 35mm Leica rangefinders he's probably holding the thing one-handed and thinking Leica lenses are doing anything for resolution.

And there are folks like the late Galen Rowell who blew up 35mm way beyond my taste, and then said things like "see, with a tripod 35mm looks just like medium format". Not to my eye. But then I really appreciate what Christopher Burkett accomplishes by not going much more than a factor of 3 or so larger than his negatives.

This is one of the reasons photography is an art; when making that image there are lots of trade-offs to be made, and most of them come down to knowing how to apply the materials you've chosen to pass the message you want.

Meuon, a clarification I wanted to make on a comment above: There are extremely short lenses that aren't fish-eye lenses. They still maintain straight lines and keep perspective. They can do weird things to perspective (when used for portraits, everyone looks like a shyster used-car salesman), but the effect isn't exactly "fish-eye".

I understand that there are 'wide angle' lenses that do not 'fish-eye', and they work well sometimes, but they can cause straight lines to look funny to me. I've seen pictures with a lot of 'fish-eye' that work, more with just a little that I liked, and wondered which way you liked it. I've been reading a bit on 'photo.net', trying to understand less emotionally what makes a good image. I know what I like when I see it... but how do I get that type of image when I don't get it by accident? How do you see it and create it when you want to? That's skill and art.

Dan--if Meuon has a Canon, you should loan him your 17-35mm zoom. :^) Have you ever considered selling that lens? It's a great one ...

Meuon's got this sweet Sony CyberShot. No 17-35/2.8 for him. Recently I've been considering just selling the whole kit and getting a CyberShot like Meuon has, or an Olympus E20.

I'll keep my comments brief. Most of what I have to say is on http://www.normankoren.com/Tutorials/MTF7.html As for Monaghan's page, I've seen some superb 8.5x11 prints from inexpensive 3 Megapixel digitals (like my son's DC4800). Pixel quality will be even finer with the 1Ds (more dynamic range), so we can expect staggering 16x20's. National Geographic will change its policies sooner or later. One interesting observation on my page: 8.6 micron pixels have about the same resolution as 35mm film (Provia scanned at 4000 dpi and sharpened). The 1Ds, at about 8.8 micron pixels (guess) is there. Now if only I could afford it...

So Norman: a question, since you've talked about it a bit. A lot of people talk about aliasing problems in the sensors, but I'm not sure that they mean what I think of when I talk about aliasing. Any idea what the sampling function per pixel really looks like? I assume the pixels don't overlap, but is it a step function? Is there some fall-off near the edge of the pixel filter? Is it a gaussian function, that, in conjunction with an optical splitter and a few more CCDs could be made to be an almost perfect filter?

If it's a square step function, then as long as the image recreation is a square step function there's no aliasing, just a limit to resolution as you try to scale up.

Dan, Most sensors with large pixels have anti-aliasing filters, which spread the response. I discuss the response in http://www.normankoren.com/Tutorials/MTF2.html#Scanners I'm using sinc^4 for the response of digital cameras with anti-aliasing and bayer arrays-- not a rectangle. More detail on http://www.normankoren.com/Tutorials/MTF7.html#Simresults I don't have enough expreience with digital cameras to comment on aliasing. I haven't seen it often, but it can be noticeable on patterns with regular textures, like fabrics or window screens. I don't think it's that serious, at least not most of the time. What have experienced digital camera users seen?

The main artifact I've noticed has been fringing on diagonal lines, I assume because of whatever technique is being used to reconstruct color from the Bayer pattern sampling. Well, that and JPEG blocking, which higher end cameras can get around, but it takes so long to do that on my application using the embedded D30 and D60 we're just using fine JPEG.

I'll have to find some mesh to take pictures of with the D60 I've got on my desk.

RE: Bracketing. What I am about to suggest may not be possible, or it may already be part of some prosumer or professional digital cameras. What I'd like to see is selectable bracketing where the actual "exposure" is made only one time but, for sake of example (3) separate images are stored into three separate memory buffers based on modified (bracketed) output to each from the CCD sensor. That way, you'd get perfect pixel-for-pixel registration of the resulting images, which might make any subsequent masking with a tool like Photoshop easier?

One caveat, it would seem that using this approach you wouldn't be able to take advantage of "turning up" or "turning down" the sensitivity of the pixels in the image sensor (i.e. changing the "exposure"), thereby preventing the capture of additional image dynamic range.