Camera Design Wish List

[richard_shewmaker1]

<p>We all want better digital cameras, but we all have different ideas of what "better" might be. Keeping our dreams within the limits of physical laws and current technology, let's share ideas and fantasies.<br>

My first one will seem a little weird to most digital photographers. I want a point-n-shoot camera dedicated to black and white photography! The sensors in digital cameras create color by masking segments of each pixel with filters for red, green, and blue, then combining those colors in the camera's computer to assign a color for the whole pixel. My dream B&W camera would do away with the filters, thus increasing the sensitivity by allowing more photons to reach the light sensors. The effective resolution would also be increased by a factor of six (the number of segments combined to make a color pixel). This camera would be no more expensive to manufacture than the color version. In fact it may be cheaper! A failed segment in a color sensor destroys the entire six-segment pixel, while in my B&W version lose only a tiny negligible speck meaning a much higher yield rate for each silicon disc. Why P&S and not a B&W D3 or 1D Mark III? Marketing. Nobody is going to pay thousands for a dedicated B&W camera. But I would gladly spend a few hundred for a "G11b".<br>

Now what's your fantasy?</p>

[miserere_mei]

<p>Richard, I wrote about my (realistic) dreams for DSLRs <a href="http://enticingthelight.wordpress.com/2009/05/26/the-time-has-come-for-a-new-dslr-paradigm/">here.</a> In regards to your B&W P&S, I actually like the idea; a bit like those people who spend $150 turning an old P&S into an IR camera--it might not be worth spending the money to do that on a DSLR, but it's worth it on a P&S.</p>

<p>However, I'm wondering where you got that number 6 from in regards to increasing resolution... Further, each pixel's colour is determined from the information of 4 pixels, not 6. <a href="http://www.cambridgeincolour.com/tutorials/camera-sensors.htm">Read here.</a></p>

[richard_shewmaker1]

<p>I stand corrected. Even so, my camera would have four times as many effective pixels. And it would work at higher ISO's. I wonder what the filter factor of those built-in filters is.</p>

[godfrey]

<p>This "improved resolution" is a myth, Richard, or at least an inaccurately expressed part-truth.</p>

<p>Bayer matrix sensor digital cameras have the full spatial resolution of the effective area in the photosite array; their *color* resolution is approximately 1/4 the total number of photosites since they interpolate the RGB color mix for each pixel position by looking at the pixels surrounding each pixel position in overlapping arrays as the article pointed to above shows. Fovean sensor digital cameras have equal spatial and color resolution since each photosite position has red, green and blue sensitive components, but they typically have lower spatial resolution than a Bayer matrix sensor camera.</p>

<p>The problem with a monochrome-only sensor is that you will not record any color information: the spectral response is fixed to the chemistry of the sensor, just like it is with B&W film. B&W rendering often depends upon being able to differentiate equal intensity but different color values to separate gray scale tonalities. Without capture of full RGB color information, you will have to use traditional B&W filtering approaches to obtain gray scale tonal separation, which is a much less flexible way to work than to capture full color information and use channel mixing, in one form or another, to fit the spectral response scale to your scene as desired.</p>

<p>My camera design wish list is much simpler. I see little wrong with most of the cameras I have now other than that they have more features and are more complex to exploit fully than I need, and often have clumsy ergonomics as a result. I'd like to see options for cameras with fewer features and better, more ergonomic and more sensibly laid out basic controls ... at prices less than the Leica M8 ...</p>

[joseph_wisniewski]

<blockquote>

<p>The effective resolution would also be increased by a factor of six (the number of segments combined to make a color pixel).</p>

</blockquote>

<p>Actually, the effective resolution is increased by a factor of about 2 (in megapixels) or 1.41 (in terms of image size. This is borne out by my experiments with a monochrome converted D100 (not IR, real monochrome, dissolving the Bayer filters) and Iliah Borg's experiments on a monochrome D2X.</p>

<blockquote>

<p>This camera would be no more expensive to manufacture than the color version. In fact it may be cheaper!</p>

</blockquote>

<p>It ends up quite a bit more expensive. People would only order pure monochrome cameras in, at best, 1/100 the quantity of the color ones. That makes the special run of sensors, boards, firmware, manuals, etc. a lot more expensive. It also means the camera doesn't go through normal distribution networks, doesn't get stocked in as regular merchandise in stores, and there's likely to be little to no competition.</p>

<blockquote>

<p>A failed segment in a color sensor destroys the entire six-segment pixel, while in my B&W version lose only a tiny negligible speck meaning a much higher yield rate for each silicon disc.</p>

</blockquote>

<p>A typical sensor had dozens (and sometimes hundreds) of "dead" (black) or "stuck" (full red, green, or blue) pixels "mapped out", that is, placed in a table in the camera's flash memory that tells the camera how to get rid of them.</p>

<blockquote>

<p>Why P&S and not a B&W D3 or 1D Mark III? Marketing.</p>

</blockquote>

<p>Nope. I looked into this already. Monochrome conversion is virtually useless on a P&S. The cameras already operate so close to their diffraction limits (especially those silly 10-15mp P&S cameras) that the resolution gain from going monochrome is near non-existant. The small sensors don't lend themselves to the kind of wide dynamic ranges one likes to see on a monochrome image, and the lenses aren't known for contrast or well preserved shadow detail</p>

<blockquote>

<p>Nobody is going to pay thousands for a dedicated B&W camera.</p>

</blockquote>

<p>A manufacturer could offer one for a premium of just a few hundred over a color version. So you could see things like a $1400 Nikon D90m or a $1500 Canon 50Dm. If a manufacturer offered one, it would still have microlenses, and the base ISO would be around 800, with usable ISO 6400.</p>

<blockquote>

<p>But I would gladly spend a few hundred for a "G11b".</p>

</blockquote>

<p>I'm not sure how a P&S that costs more than an entry level DSLR like a rebel ties into your arguments about price. The dRebelM (dREBELm?) would not have the diffraction problems of a G11m.</p>

<p>If you want to ship me a G11, I'll take a try at a G11m. But I have to warn you it's a high risk procedure, especially on a Canon. You may end up getting a dead sensored camera back. The only known Canon DSLR attempt, a 20D (not one of mine or Iliah's) met with failure. The Canon 20D color filters proved resistant to all common aromics and aliaphatics.</p>

[godfrey]

<p><em>.."Actually, the effective resolution is increased by a factor of about 2 (in megapixels) or 1.41 (in terms of image size. This is borne out by my experiments with a monochrome converted D100 (not IR, real monochrome, dissolving the Bayer filters) and Iliah Borg's experiments on a monochrome D2X..."</em> <br /><br /> I'd be interested to know what you mean by "effective resolution" and the details of your monochrome conversion. That is, did you replace/remove the antialiasing filter, how did you process the images captured in that evaluation, how did you measure the resolution? <br /><br /> t'anx!</p>

[joseph_wisniewski]

<blockquote>

<p>Further, each pixel's colour is determined from the information of 4 pixels, not 6.</p>

</blockquote>

<p>Actually, in <a href="http://en.wikipedia.org/wiki/Demosaicing">common algorithms</a> , a "final" pixel can be computed from as many as 13 "Bayer" pixels, the local one and as many as 12 neighbors, depending on the algoritum used for the Bayer demultiplex. That's the kind of thing you see in good raw converters like P1C1, ACR, Lightroom, or NC. There are high order interpolators like Lanczos that may use a square of 64 pixels to compute each pixel, and "frequency domain" algorithms that may use thousands. No one ever (not in the last 20 years) used anything at all like what is described in that Cambridge In Color article. (I'm used to a much higher quality of article from their website: the one you pointed out is, basically, tripe. This is unusual).</p>

<blockquote>

<p>The problem with a monochrome-only sensor ... you will have to use traditional B&W filtering approaches to obtain gray scale tonal separation, which is a much less flexible way to work than to capture full color information and use channel mixing, in one form or another, to fit the spectral response scale to your scene as desired. (elipses mine)</p>

</blockquote>

<p>Actually, being one of the few people here to have a real monochrome DLSR (with two more in the queue waiting for better tooling) I'd say that the traditional filter method has some really amazing advantages.</p>

<ul>

<li>You can have any filtering: red, orange, yellow, green, blue, violet, ultraviolet, infrared (in several different pass bands). I know that some B&W conversion software has sliders for "tweener" colors like orange or violet, but in the end, there's really ony red, green, and blue, and the red is wide enough so that it looks nothing like a traditional R25 or R29.</li>

<li>Many lenses have chromatic aberration problems. Although software can correct lateral CA (but not longitudinal) this results in a resolution loss (and artifacts, due to the dramatic effect of even 1/2 pixel width of CA on the Bayer demultiplex algorithm). A filter on a monochrome sensor can completely wipe out CA. (relatively narrow filters like red 25, green X1, and blue 47 are especially good at this).</li>

<li>As I mentioned earlier, you really do get about twice the resolution of a color sensor. No AA filter (monochrome aliasing artifacts are not as visually distracting as color moire, and if you really are aiming for pixel pitch resolutions, diffraction can often provide all the AA filter you need to kill moire in certain critical applications).</li>

</ul>

[joseph_wisniewski]

<p>My wishlist item simply isn't going to happen. I'd like to see a camera where much of the firmware is open and accessible. Look at some of the things hackers figured out how to do to Canon DSLRs, using a system that isn't really "open" but is "hackable". Picture what one could do if a lot more of the camera's firmware were as open as a cell phone using Android or Javascript (like my Moto Zine, which I've already written simple applications for).</p>

<p>I'm confused by Godfrey's desire for "better, more ergonomic and more sensibly laid out basic controls". Most P&S cameras are horrid, but the typical DSLR has incredibly good ergonomics, compared to any mechanical camera (or fake mechanical, like the Leica M8).</p>

[thomas_sullivan]

<p>my wish for advancement in digital cameras, P&S, DSLR, digital backs, etc is to have ALL camera manufacturers standardize the format for the Video they are all seemingly tending towards. Not that I shoot much video in my P&S, I have had 3 so far, and not one of them could use the same program for editting.....or even just playing. From what I've seen of DSLRs this trend has continued. This is totally unacceptable.</p>

[joseph_wisniewski]

<blockquote>

<p>I'd be interested to know what you mean by "effective resolution" and the details of your monochrome conversion. That is, did you replace/remove the antialiasing filter,</p>

</blockquote>

<p>The experimental D100m has no AA filter. The entire filter stack (AA, blue/green "equalizing" filter, and IR blocker) goes away. The sensor is breached, the microlenses and color filters are dissolved by solvents, the sensor is cleaned and gets a new quartz cover glass. Working on argon fill. I intend to keep that same methodology on the production D90m. I've noticed that monochrome aliasing is much less of a problem than it is on color cameras with AA filters removed.</p>

<blockquote>

<p>how did you process the images captured in that evaluation,</p>

</blockquote>

<p>Color images are processed using a late version of dcraw and the AHD (adaptive homogeneity-directed) algorithm, and "sanity checked" against Lightroom. Monochrome images are processed using software that level equalizes and fourth order linearizes the channel mismatches (there are some, due to firmware on single channel A/D cameras like D100, and also hardware gain differences between channels on cameras like D2X and D3 with 4 or 8 channels).</p>

<blockquote>

<p>how did you measure the resolution?</p>

</blockquote>

<p>Pseudorandom test pattern and a homomorphic deconvolution algorithm to compute MTF in the frequency domain. I can also get really wonderful plots of the point spread function of pretty much any camera and lens this way.</p>

[miserere_mei]

<p>Joseph, I'm fascinated by your posts... Have you tried converting any Pentax DSLRs? I have a K100D Super that is just sitting on the shelf right now and I'm so tempted.... :-)</p>

[godfrey]

<p>Excellent, thanks Joseph. I appreciate you taking the time to describe it. Your approach is much more sophisticated than the "remove the Bayer filter mosaic" mechanism to produce a monochrome only camera suggests, which is what I was responding to. I certainly believe a 1.4 to 2x resolution improvement without the AA filter and other filter pack elements. </p>

<p>Traditional filter methods for B&W photos worked for 180 years, they should work now. They're very good, as you said, but a heck of a lot more business to deal with than most people who just want to 'dumb down' an RGB camera to B&W usually know about. </p>

[j.w.]

<p>Well, a monochrome sensor could be developed that has a bandwidth from IR to UV; then a filter over the lens could work just like in the good old days of monochrome films. Sorta like old-school channel mixer. ;)</p>

<p>Given the economics of scale, it's obviously more efficient to just use a color camera and do B/W conversion in post. But I wonder if the larger photo-sites of a monochrome sensor would give great low-light capability? That, it seems to me, would be its primary advantage.</p>

<p>~Joe</p>

[joseph_wisniewski]

<p>Godfrey, you're quite welcome. Indeed, monochrome digital has all sorts of nuances and outright "gotchas" that make it quite a bit more complex than it seems at first glance.</p>

<p>Joe VanCleave, a monochrome camera does great at low light levels. Even my monochrome D100, which loses a stop or two because the microlenses were dissolved in the same operation that dissolved the color filters, is a stellar low light performer, producing results at ISO 1600 that are well beyond what a B&W conversion of a color D100 image looks like. I'm so looking forward to seeing what the monochrome D90 can do. I'm expecting that camera to really torture lenses.</p>

[godfrey]

<p>@Joseph ... regards my wish list of wanting <em>"better, more ergonomic and more sensibly laid out basic controls"</em> ... </p>

<p>While most DSLRs today are decently laid out, they often suffer from what is to me too much feature overload.</p>

<p>Probably my own sense of aesthetics ... I like simplicity: that M8 which you refer to as a 'fake mechanical camera' appeals to me a lot for its stark simplicity, which I'd be very comfortable with. The Olympus E-1 I use occasionally seems much simpler control-wise than most of my other DSLRs and is a very appealing camera on that basis. Configured as I want it in the menus, the basic layout and control organization is easy to learn, easy to remember and lets me concentrate on exposure, focusing, framing and being there with the subject, with few gotchas to look out for.</p>

[tom_anderson9]

What I want would be easy to do, but they'll never do it. I want the exact opposite of a modern p&s digital, in digital:<p>

The size of a minox or smaller<br>

No LCD at all, small viewfinder<br>

12x18mm sensor, 3mp or so, optimized for high ISO<br>

No zoom - fixed focal length lens around 35mm eq<br>

Either a setting for 'hyperfocal' with fixed aperture and focus, or fixed focus/aperture<br>

No flash, or detachable flash<p>

Basically, something I can carry that will be tiny for snapshots. There is a market for this, as demonstrated by cell phone cameras. You don't need zoom and autofocus for a point and shoot camera, and number of pixels is far down the list of desired features, especially compared to low light performance.