Wednesday, March 31, 2010

Kodak Officially Announces W-RGB CCDs for HD Video

An official PR on Kodak site announces that it is deploying the TRUESENSE W-RGB Color Filter on the 1080p/60fps format KODAK KAI-02150 CCD, which is said to enable "the capture of color images with the sensitivity typically associated with monochrome cameras... users can realize a 2x to 4x increase in light sensitivity (from one to two photographic stops) compared to a standard Bayer color sensor". Kodak also sent me a pdf file with high-resolution comparison pictures that support this claim (click on pictures to expand):


Kodak is also expanding its HD image sensor portfolio with addition of a new 1/2-inch 720p/130fps format CCD based on 5.5um pixels - also available with the TRUESENSE Color Filter Pattern. In addition, both the 720p and 1080p image sensors will be available in a smaller, CLCC package configuration to enable development of more compact camera designs.

To assist manufacturers in modifying their camera software to work with this new color filter pattern, Kodak has developed a set of tools that document the new image path used by sensors that deploy this technology. Providing both an overview as well as a practical implementation of the software algorithms used, these tools can be used not only to modify an existing image path to work with the new color filter pattern data, but also to compare the results of that implementation to a reference image path provided by Kodak.

Engineering Grade devices of the KAI-02150 with the KODAK TRUESENSE Color Filter Pattern are available today, and Standard Grade devices are scheduled to be available in Q2, 2010. The KAI-01150 will be available in monochrome, Bayer Color, and KODAK TRUESENSE Color Filter Pattern configurations. Engineering Grade devices of the KAI-01150 are scheduled to be available beginning in Q4, 2010, with Standard Grade devices scheduled to be available Q1, 2011.

6 comments:

  1. I want whoever took these photographs to take all my pictures for me : they are brilliant ! They managed to capture two exactly identical images (traffic lights, positions of the cars, even the water in the fountain in the background) with supposedly two cameras each containing a sensor with a different CFA painted on.

    ReplyDelete
  2. The images are a bit different, taken from a bit different locations

    ReplyDelete
  3. Not as difficult as you may think. 2 cameras side by side(if you look close there is parallax), capture multiple frames, then pick the 2 frames that match the closest.

    ReplyDelete
  4. yes... looking at it a little closer I guess there is bit of parallax and the dust particles are not at the same positions. I may have gone a little too fast suspecting a photoshop job.

    ReplyDelete
  5. I having trouble with the claim of 4x gain in sensitivity via CFA pattern changes. Let's say the filtered pixels had a sensitivity of Q0, and the clear pixels had a sensitivity of Q1. If Q1=4Q0 and we replaced ALL the pixels with clear pixels then we could gain 4x in sensitivity but then there would be no color. I am thinking Q1/Q0 must be about a factor of 3, at best, so even a "monochrome" (panachrome?) sensor would have a factor of only 3x improvement. So either Q1/Q0>>3, or something else is going on like increased NIR sensitivity or other pixel QE improvements.

    2x improvement, as illustrated in the photos, can be explained with 50% clear pixels and Q1/Q0=3.

    Maybe Mr. Kodak can explain the 4x claim. Also, it would be good if they showed also a 1.4x improvement in YSNR rather than a 2x shifted histogram. Histogram gains are easy compared to YSNR improvement.

    ReplyDelete
  6. You can trade resolution for sensitivity. This is not fair, but you can easily justify 4x gain in sensitivity by opening up the kernel size and erasing away some effective MTF, especially that MTF related to certain spectral content. Engineering trade offs and marketing have collided and marketing rectified that trade off to expose only the benefit. Some might believe that the trade off is a good one. I doubt that they use NIR improvements but that is a good thought since you are trying come up with ways this claim could possibly be true. I only doubt this because, you may need to filter out the IR out in the color channels before the photons are captured. Otherwise removing the IR content after capture would increase the noise. It is not easy to selectively filter the IR at individual filter tiles before capture.

    ReplyDelete

All comments are moderated to avoid spam.