Since all of the big Photokina camera announcements seem to be out (Canon 1Ds, Nikon D2X, Fuji S3, and a couple thousand point-and-shoots), I figure it’s time for me to post my list of what I’d like to see the camera industry provide. I’ve been thinking about most of these for years. In that time, I’ve seen new cameras come and go, but I haven’t seen a whole lot of real innovation, particularly in the DSLR space, where all of the manufacturer’s effort has been focused on image size and speed.

None of these ideas are mind-bogglingly fantastic; some of them are admittedly a bit marginal. Some of them may well be bad ideas–I’d be amazed if at least some of these haven’t been tried out in the labs and found wanting. I haven’t seen any of these discussed widely online, though, so I figured I should probably share them.### In-body multiflash support.

I’d like to see in-body multiflash support. Canon and Nikon’s high-end flash systems support master/slave operation, with support for multiple flash units. Once unit goes onto the camera itself, and the others can be set up anywhere nearby, as long as they have a good view of the on-camera flash. It’s a nice concept, but it’s never worked very well for me in practice. Part of the problem is that you need multiple $300 flashes to make it work.

What I’d like to see is support for using the camera’s built-in flash as the master. You could control the flash system using the camera’s LCD instead of the small nasty LCD controls on the flashes. The big problem would be the angle of coverage of the built-in flash, but there are some easy fixes for that, too.

None of Canon’s higher-end bodies have ever included a flash, largely because pros don’t have much use for small flashes, but partially because pop-up flashes aren’t very rugged and they’re hard to weather seal. If you’re just planning on using the built-in flash as a multiflash controller, then you can get around all of this. Just make it IR-only, don’t include a pop up, and make the entire top of the prism hump IR-transparent. Just shoot the flash straight up. Since there are no moving parts, it’s easy to seal and fairly rugged. Since it has 360° coverage, it should work great indoors. And since it’s IR, it won’t interfere with your other lighting.

Bluetooth

I like Bluetooth–I have a Bluetooth phone, a headset, a GPS receiver, and a USB Bluetooth dongle for my Mac. It works great as a low-bandwidth, low-power communications mechanism. I’d love to see it on a camera.

What it’s not: * a way to download pictures. It’s way too slow. It makes USB 1 look dangerously fast.

What is is: * A wireless way to use GPSes with the camera. Nikon supports serial GPS units, but then you’re suffering from dangling wires. With a bluetooth GPS, it’d be easy to annotate each frame with the location that it was shot. * An alternative to on-camera microphones for voice annotation. Better weather sealing on the camera. * A remote-control system. You could set exposure and trigger the shutter wirelessly from a laptop, PDA, or even phone. * An improved system for multiflash systems. Instead of using optical signals, the body could talk to the flashes using Bluetooth. This would let the body query the flashes directly, one at a time, allow a lot more flexibility, and cut down on the need to have a line of sight between the master flash and each slave.

Of course, you could use nearly any cheap, low-power wireless standard for any of these. Bluetooth has the advantage of being available on the market today and being fairly easy to work with.

Floating-point image formats

Dynamic range is the last big hurdle facing digital photography. Traditional film media can handle a much wider range of light and dark then digital image sensors. Black and white film is traditionally viewed as recording 10 stops of contrast; color negative film is around 7 stops. Color slides and digital cameras are usually closer to 4 or 5 stops of usable contrast. The eye, unaided, is good for well over a dozen stops.

There are two sides to this problem. First, current imaging chips aren’t very good at handling especially bright light, and they tend to clip the brightest parts of the image off. The Nikon D2X and the Fuji S3 both have technologies designed to combat this.

However, even with an ideal sensor, current cameras are limited to 7 or 8 stops, simply because they use 8-bit JPEGs to store their images. Unless you’re willing to use RAW mode on the camera and spend a couple minutes tweaking each image and then dump it into Photoshop in 48-bit color, eating 30–80 MB per image, you aren’t going to go beyond 8 stops.

The pain doesn’t stop once you get the image into Photoshop, though. The problem is that the normal representation for color images in cameras and PCs uses one fixed-size integer per color. Typically, you have either 8 or 16 bits of information per color, with 3 colors per pixel, for a total of 24 or 48 bits of data. The problem is that the information that you care about (lightness vs. darkness) isn’t spread over those bits in any sort of optimal pattern. With an 8-bit integer, the brightest stop of light uses values from 128–255, or half of the total range. The next-brightest stop goes from 64&ndash127, then 32&ndasah-63 for the third-brightest stop. After peeling off the three brightest stops, there are only 32 color levels left for the remaining 3 or 4 stops. If you do much processing at all, you’re going to find banding and noise lurking in the shadows.

There’s an easy way around this, and movie special-effects types have been using it for years. Instead of using integers for color values, use floating-point numbers. Floating point is what computers generally use whenever they’re dealing with real-world numbers, like measurements or sizes. Generally, any time you see a decimal point on a computer, you’re using floating point numbers. Internally, they’re stored in the computer in a form kind of like this: M * 2^E. The computer keeps track of M and E (technically known as the mantissa and the exponent), and the rest of it is just implied. So, for example, a 32-bit integer can represent numbers from 0 to roughly 4 billion. A 32-bit floating point number can represent numbers from -10^38–10^38, but at the cost of a bit of accuracy. Fortunately for us, the accuracy comes where we care about it the least–in the lower-order bits where sensor noise lies.

Instead of using a 32-bit float for each color, we can even cheat a bit. We could get by with a 16-bit float for each color, with an 11-bit mantissa and a 5-bit exponent; that’d be enough to cover 32 stops with nearly as much color detail as modern sensors can record for their brightest stop. Or, we could cheat and share the exponent between the three colors; done this way, we could fit three 14-bit mantissas and a 6-bit exponent (64 stops) into 48 bits, or three 9-bit mantissas and a 5-bit exponent (32 stops) into 32 bits.

The advantage of any of these formats is that they’d hold the same amount of detail in each stop, rather then bunching it all up in the brightest bit of the image. This would allow a number of small improvements and one very large one: we could finally represent colors brighter then white. Even if the screen or printer can’t reproduce a specific bright color, you can still represent it in the image.

(Apparently Apple has some support for this in Tiger, and nVidia’s newer graphics cards can do 16-bit float displays)

Built-in hard drive

I’d love to have a DSLR with a fixed 1.8 inch (iPod-sized) hard drive instead of a CF slot. You can easily get 40 GB drives in that form-factor today, with 60 GB drives on the horizon. A 40 GB drive would hold almost 3,000 1Ds mk II raw images, or over 10,000 D60-sized JPEGs. In other words, you could shoot almost any event without needing to worry about storage. Just shoot, and sort it all out later. No card swaps, no dropping them in the mud, no dust in the CF slot, just shooting.

If you drop the camera, the drive might have problems, but I’ve dropped my iPod from waist level without it croaking. What would a 3-foot drop onto a hard surface do to most camera bodies? Drive failure could be a problem, but it’s probably no less likely then shutter problems or stuck apertures, and people have dealt with those for decades: you bring more then one camera.

If you’re worried about crashes eating the images that you’ve already shot, then either keep an assistant handy to do downloads when you swap bodies, or use a wireless adapter like Nikon or Canon’s and have it download images onto a laptop while you’re shooting. The in-camera drive just turns into a big buffer and a backup for your laptop’s copy of the images. Done correctly, you’d end up with two copies of every file in fairly short order; try that with CF cards, and compare that to the dear lab please don’t eat my film this time fear that we all used to have.

That doesn’t even touch the performance factor–the specs for Toshiba’s 40 GB drive suggest that it should be able to sustain around 20 MB/sec on the outer edge of the drive. Compare that to 5 MB/sec for CF cards or 7 MB/sec for SD cards. Shooting 15 MB RAW images (1Ds mk II-sized), that’s 1.3 fps sustained. Shooting 5 MB JPEGs, that’s around 4 FPS. With a camera like Canon’s 20D, you’d be able to shoot 1,000-JPEG bursts.

Of course, it’d have a big battery hit, but I don’t see that as a big problem right now. The latest generation of pro cameras can shoot over 1,000 frames per battery, and battery technology and low-power electronics are boosting that around 50% per generation. We’ve reached the point where you can shoot 15 GB of data per battery swap, but you have to go through a dozen or so flash cards. Trading battery life for storage capacity doesn’t seem like a bad trade-off to me.

Digital Rangefinder Camera

I’d love to see a small, interchangeable-lens digital rangefinder camera. Like film rangefinders (Leica, etc), it can be tiny because it doesn’t need a SLR’s mirror. Unlike film rangefinders, you’d have the options of doing autofocus using the image chip for AF. Unlike digital point-and-shoots, you’d still be able to focus manually, even in low light, which is where rangefinders have always excelled. And unlike film rangefinders, you’d be able to see through the lens if you feel the need–think long lenses or parallax. It be a great mixture.

This would be a great market for one of the smaller manufacturers, like Olympus, Pentax, or Minolta. They really can’t compete with Canon or Nikon with DSLRs right now, but they can produce a smaller body that takes small lenses and then produce an adapter that will let them use some 35mm SLR lenses with the body. Its a niche, but it’s one that Canon and Nikon can’t move into without cannibalizing their DSLR market, so it’s probably fairly safe.

Epson has started down this road with their R-D1, but it takes Leica lenses, so it can’t do autofocus. To do this right, you’d really need a new lens mount with electronic controls and a really short focus distance; to the best of my knowledge, no current lens mount combines the two. I had hopes for the 43 people, but they don’t seem to be headed down this road (or anywhere else, either).