Kodak Moment: Bring in Da Light, Bring in Da Photons

We think that this Kodak RGBW pattern put more weight on luminance than chrominance with more white mosaics ! On the contrary, our company proposes a more balance VPW RGBW pattern, please check www.vp-dynamics.com/vparticles/vpw.pdf and www.vp-dynamics.com

heyyyyy abeeeey:]

Sensitivities are getting better, but there is a very long way to got. High-end night surveilence cameras use photon-multiplier techniques so that good images can be seen by star light. Anyone want to guess when we see a Kodak digital camera with 0.01 lux sensitivity?

Frequently seen in this business: "Everything old is new again."

It's interesting to me to see a return to separate luminance (RBGY) in video pickups. Early broadcast cameras (like the RCA TK42/43 even prototype TK44, GE and Marconi) and the industrial Sony DXC-5000 used separate luminance channels. These were abandoned for S/N and chroma fidelity (Livingston color error) reasons so currently RBG is standard.

Oh yeah, Wikipedia says I forgot to mention Sony's RGBE (red, green, blue, emerald) as one of the 4-color filter arrays that preceded Kodak's RGBW announcement.

To Uttam Sahu: Your comment regarding the need for more processing power is spot on. Kodak has stated in its release materials that this new color filter array will probably require more procesing power than do cameras with Bayer filters. However, consider these three facts: 1. Every jump in pixel count from 2 to 10 Mpixels has also required more processing power. 2. The cost of the processor in the camera is truly insignificant compared to the sensor, the lens, or the battery. The image processor generally occupies a relatively small part of one chip. 3. The purpose here is to produce better pictures with more detail. If the RGBW filter fails to deliver this improvement, then it will not be worth the change and it will not replace the Bayer filter. Other 4-color filters that preceded this Kodak announcement (JVC's CMYW and Canon's CMGY filters for example) did not change the market, possibly because they did not deliver a better image. The vast majority of digital cameras appear to still use Bayer filter arrays. What's probably more important is that Bayer algorithm work has been going on for 30 years, so the algorithms are mature. I doubt that the RGBW filter algorithms are nearly as mature, but they will no doubt only take a year or two to improve to current Bayer-filter algorithm levels because they can stand on the existing knowledge developed during the Bayer algorithm development work.

What about the performance in reducing aliasing artifacts like color aberrations and moire patterns? As the color filters are at more distance from each other, a complex algorithm might be required to compute the exact color, adding to cost of the system in terms of MIPS and/or power.

To my taste, which is scientific camera demosaising algorithm provider, this new RGBW matrix shows 2/16 red, 2/16 blue, 4/16 green and 1/2 white pixels, and my stomach feeling is fine with luminosity information, but pretty loose with colors. Above is just first impression.

I thought that some cameras, such as the Canon G1 are using a Color Filter Array (CFA) containing a 4 color CMGY pattern

Will this be a similar story as in the LCD-panel industry? The idea of using RGBW pixels to improve rendering and/or efficiency (of a display ? hence of an image sensor) is indeed not new (see for example www.clairvoyante.com). The biggest hurdle for a success story is, as usual, the question of cost, cost and cost. If the new filters can be manufactured at lower cost and be made compatible/transparent to existing standards, then end customers might be willing to buy improved performance sensors (at the same low-price ... of course).

It's true that the Foveon sensor doesn't use a Bayer filter, but not true that the Foveon sensor doesn't filter the light. The Foveon sensor uses a stacked arrangement of semi-transparent sensors with blue on the top, at the surface, then green, then red. So the red light has to make it all the way through the blue and green sensor levels to get to the red sensor zone. Wikipedia has a nice explanation: look up Foveon X3 sensor.

What is the relationship between this technique and Foveon's invention if any?

Or just use the Foveon sensor that doesn''t use a filter at all.

Human eye has only 6 mln chrominance sensors (cones) and 100 mln luminance sensors (rods). My feeling is that it make sense to emulate nature to improve digital camera picture quality. Kodak is moving in that direction.

I've always noticed the characteristic color fringe in a lot of digital pictures (mine included). Looking at the pattern of the new sensor, I'm curious how algorithms could overcome what seems an inclination to inclined aliasing in this new color-filter array. That said, few (newer) folks realize that Kodak has done a whopping amount of research on color, and color film for normal snaps is tuned to human eyes, and human eyes only. Smart folks. Will miss their films, though. There's a lot to be said for low-tech analog systems, still.

Actually, Kodak announced multiple new filter patterns to replace Bayer, not just one new pattern. Kodak says that some patterns may be better suited for some cameras than for others. For example, a camera phone with a tiny (e.g., noisy) sensor might be better served by a higher-luminance filter that passes more light, allowing shorter exposures at lower ISOs. But a DSLR with a large (e.g., quieter) sensor might be better served by a higher-chrominance filter that has better color fidelity. It would be really interesting if someone invented a dynamically reconfigurable color filter that could be manually adjusted by the user or automatically adjusted by the camera, to match different subject matter -- much like white-balance settings today.