Brian's Brain

Some of you may have already seen contributed blogger Steve Leibson's writeup from yesterday on Kodak's next-generation image sensor filter array. Kodak pre-briefed me on the announcement two weeks (and a day) ago, and Liebson's comments have motivated me to write sooner than my overflowing editorial plate might have otherwise allowed. I agree with his conclusion that "it finds something useful to do with all those extra megapixels we’re getting from CCD and CMOS sensors these days." Then again, those of you who've already read my imaging feature article from mid-March, along with its online addendums, probably aren't surprised at this stance.

However, unlike Leibson, I personally don't quite think Kodak's "just revolutionized digital imaging." For one thing, this was a technology announcement; Kodak doesn't expect the first sensor based on the panchromatic-appended approach to be sampling until at least some time in Q1 of next year. For another thing, the technology's frankly not new. JVC's GR-HD1 and JY-HD10 first-generation HDV camcorders, introduced in early 2003, "incorporate a hybrid complementary/primary matrix of clear, green, cyan, and yellow filters", as I wrote in a mid-September 2004 feature article. Granted, JVC's video-tuned sensors use an odd mix of color filters, whereas Kodak's filter array proposal employs conventional RGB primaries. But they both append the patterns with panchromatic (i.e. clear) filters to boost luminance.

I pointed out JVC's seeming 'prior art' to Kodak during my telephone briefing with the company's image sensor marketing manager, Mike DeLuca. Two weeks (and numerous reminders) later, I finally got this response on Wednesday morning:

We recognize that the use of clear pixels in combination with other colors is not unknown in sensor design. In fact, Kodak cited a number of patents as prior art in our applications. However, there are some important differences to point out concerning Kodak's new technology. For example, the JVC design replaces a clear pixel for a single element in a 2x2 array, so that 25% of the pixels are panchromatic. Kodak's designs use at least 50% panchromatic pixels, allowing the panchromatic pixels to serve as the luminance backbone of the image processing chain (much as the green channel in a Bayer patter serves as the luminance backbone for processing the Bayer pattern).

In addition, the specific patterns used by Kodak have an advantageous arrangement of color pixels that allows combining nearby like colors together in order to overcome the photographic speed deficit of the color pixels vs the pan pixels. This "speed matching" is an important component of the underlying science behind how these patterns work, and the interpolation algorithms that work with our patterns use these combined color pixels as an integral part of the processing chain. Finally, the specific imaging path used with these new Kodak patterns - a panchromatic path at full resolution and a color path at lower resolution - appears to differ substantially from the JVC interpolation used in this device.

Leibson touched on my final motivation for pragmatism when he wrote, "unfortunately, this superior filter array pattern can’t just be dropped into existing camera platform designs. Algorithms created to use Bayer RGB information are wholly unprepared to accept RGBW data. Camera vendors will be scrambling to adapt to this new world. Image-algorithm writers have just had their reset buttons pressed." He's right. Therefore my conclusion that there'll be little motivation for high-volume potential customers to standardize on sensors based on Kodak's technology, unless the Rochester, NY firm licenses a meaningful alternate-source sensor supplier.

Granted, if the algorithms are running in software on an Analog Devices, Texas Instruments or other vendor's processor, they'll probably be more rapidly ported and available than if Kodak is forced to wait for support from hardwired image processing 'engine' suppliers like NuCore Technology. But, particularly given that Kodak's a fairly niche player in today's CCD and CMOS sensor markets (a disproportionately large percentage of the company's overall sensor-stimulated revenue comes from patent license royalties, not from silicon sales) I just don't see any big sensor customer taking the cost and supply risk on a proprietary filter array approach.

Sorry, Steve. With all due respect, I don't concur with your optimism. What do you think, folks? Am I being too hard on Eastman Kodak, and on my blog cohort?