As shown by its latest $1,400 Camedia C-2500L digital camera for consumers, digital cameras are in dire need of more processing horsepower for image refinement after they capture an image. Without it, said Akio Terane, senior engineer of the digital imaging business development department at Olympus, users will begin to see diminishing returns no matter what how many millions of pixels the camera boasts.

"People think that as the number of pixels increases, the resolution should be better automatically," Terane said, through a translator. "But really, the depth of an image and its resolution involve a relationship between the optics and processing."

When Olympus designed its previous digital cameras, that would have been considered overkill. "When we came out with our 1400L, we had been using large RISC and CISC processors," Terane said. "The theory was to do processing for imaging using RISC and CISC and to do it in software. But we shifted that strategy, so we changed our process to get higher speeds."

Terane said he and his team set out to make a camera that could turn out the best possible digital image, whether for a print or a PC display — an image proportional to the number of pixels the camera can collect from the CCD. As elementary as this sounds, digital camera makers have not always been able to do it.

This is true when displaying on a screen digital images taken from cameras with megapixel CCD image sensors, Terane said. Digital cameras are generally capable of three resolution quality levels, but when trying to downconvert from superhigh-quality SHQ to VGA resolution, "sometimes the VGA resolution image is better than the higher-resolution image," he said.

"Theoretically, when we talk about superhigh-quality mode, which is the best, and with sensor pixels going up, this should result in a better image," he said. "Whenever you have more information, you should create a better image."

To get higher resolution and better luminance, the company knew two years ago it would have to start from scratch. For starters, Olympus decided to design its own CCD sensor, figuring that was the only way to ensure it could get a high enough pixel count and still integrate well with the rest of the surrounding circuitry. And while standard CPUs were getting smaller and cheaper, they would still be woefully inadequate to do the image processing Olympus had in mind.

When making a downconversion from SHQ for VGA resolutions, the calculations normally used are relatively simple. In determining brightness, for example, a pixel's characteristics are determined simply by taking two points of the high-resolution image and doing an average.

Olympus, however, developed its own algorithm that takes into account a much wider area of pixel information, such as proximity to a red-green-blue color filters, to determine the pattern of brightness around the pixel. It's a process analogous to using a higher sample rate when making a digital recording.

Olympus decided to hardwire its TruePic processing algorithm into a central image-processing ASIC that includes a memory controller, autoexposure, autofocus, white balance and camera control logic. Terane declined to disclose how it was done or the name of the foundry, but he said the process technology was leading-edge and the number of gates used exceptionally large. "We used the highest technology available," he said.

And it's no wonder, considering the processing load such a chip is shouldering. If the processing algorithm were to be handled in software using only a general-purpose CPU, it would take nearly half an hour instead of less than a minute to refine an image, Terane said.

Still, that didn't stop Olympus from throwing in a pair of external 32-bit CPUs. A CISC engine handles the control panel interface and key inputs, while a RISC device is used to control the picture-taking process starting from when the shutter opens to the autoexposure, autofocus and white balance.

Though using general-purpose processors for image processing was turned down in favor of a custom chip, it still made sense from a cost point of view to keep to conventional CPUs for control functions. "Size-wise, RISC and CISC are much smaller than they were for our first-generation products," Terane said.

On top of that, Olympus included a second Y/C ASIC to process color and brightness, and a separate DSP and memory for displaying the final image on an LCD or video monitor.

Such high-powered processing can only be effective if it has enough data to work with. Here too, Olympus chose its own path by designing its own megapixel CCD. As with the processing engine, Olympus started several years ago with the intention of bringing out a best-of-class device. It also wanted to make sure it could build in progressive RGB color filtering, which at the time was still hard to find in an off-the-shelf device, Terane said.

The resulting CCD has 2.5 million pixels spaced 0.5-micron apart on a slice of silicon measuring 2/3 inch. To be sure, companies like Sony Corp. are getting ready to introduce 3 million-plus CCDs into the market. But Terane said there are many other factors to consider besides number of pixels and pixel pitch.

These include fine optics. In Olympus' case, it is using a 3x optical zoom, equivalent to the 36-to 110-millimeter range of 35-mm cameras. It has seven glass lens groups with seven elements, including an aspherical lens.

And consideration must be given from the beginning to other components surrounding the CCD, including timing circuitry and a 10-bit D/A converter to suppress noise, he said.

Other features were designed to make the camera feel like a conventional 35-mm. One was including 128 Mbits of DRAM to act as a buffer memory for up to five frames before preprocessing, allowing users to take a shot without having to pause between frames. In most digital cameras, such pauses can last as long as 4 seconds.

As for external memory, there is still no single solution for add-on flash memory cards. The Camedia has slots for both SmartMedia and CompactFlash.