Digital cameras break through the resolution barrier

Digital cameras have recently reached a photography milestone in that they can now match the resolution available from conventional film. Within just the last couple of years, cameras with five million pixels and more have entered the general consumer market. Camera manufacturers offer even higher pixel counts for professional and scientific cameras.

Consumers are not yet throwing away their film, though. Recording and printing infrastructure still lags behind that available for film, and digital technology is still maturing. Not only are there a variety of memory formats vying to replace film cartridges, but complementary-metal-oxide-semiconductor (CMOS) sensors are beginning to chip away at the market share of the charge-coupled device (CCD).

Resolution breakthrough

The resolution advance in digital cameras was dramatically revealed late last year, when Foveon Corp. (Santa Clara, CA) and National Semiconductor Corp. (Santa Clara, CA) announced the development of a 4096 × 4096-pixel CMOS detector chip for use in high-end digital cameras in as early as 2002. Simultaneously, Kodak (Rochester, NY) announced a 4096 × 4096-pixel CCD chip. Such chips offer more than 16 Mpixel/chip, matching the resolution of the highest-quality photographic film. Even a 20 × 20-in. print with 16 Mpixels would have no visible pixels at normal reading distance.

A few months following the Foveon and National Semiconductor announcement, Japan-based Minolta announced a consumer-oriented digital camera with a 5.2-Mpixel CCD chip with a list price below $700 (see Fig. 1). This was just one of a slew of new digital cameras with Mpixel-level counts.

The significance of such high resolution is illustrated by the results of research sponsored by Japan-based Sony Corp.¹ During the project, photography students were asked to compare digital prints with varying number of pixels against prints of the same subjects made by conventional film photography. For standard 4 × 6-in. prints, the researchers found that, overall, students considered 3-Mpixel digital images as good as ordinary photographs. The same was true for 5 × 8-in. prints and 5-Mpixel images. At least for the print sizes commonly used by the average photographer, the Sony results indicate that digital resolution matches that of film.

These advances in resolution are impacting the design of other types of digital cameras as well. An international team of engineers from the Japan Broadcasting Corp. (Tokyo, Japan) and Dallas Inc. (Waterloo, Canada), for example, has developed an experimental video camera based on an 8-Mpixel CCD.² With 2250 lines per frame and 4400 samples per line, the resolution is twice as fine as that of high-definition television.

At the moment, purchasing a digital camera with the optical quality of a film camera can still be costly. Digital cameras with large and interchangeable lenses, as is standard on 35-mm film cameras, can cost thousands of dollars. In part, this is because only the physically largest CCD or CMOS sensor can cover the image produced by a larger lens. The lens for a 35-mm camera covers an area of 24 × 36 mm, yet most CCD chips are no larger than 20 × 20 mm.

With such resolution levels come huge masses of data and data reduction challenges. A typical four-night observing run can produce hundreds of gigabytes of raw data.

CMOS in the race

While CCDs still dominate the digital camera market, the Foveon announcement was a warning that CMOS chips were becoming competitive, starting with the high-end professional cameras. These chips have the potential to be far cheaper to manufacture than CCDs. They can integrate multiple functions—detection, amplification, and transmission—onto a single chip, while consuming far less power (see Laser Focus World, March 2001, p. 147).

In CMOS detectors, individual photodetectors similar to single CCD pixels are integrated with individual transistors or groups of transistors. This allows the array to be read out over wires like random access memory, rather than by shifting charges repeatedly from one pixel to the next, as in a CCD.

Another CMOS advantage involves the capability to program chips to radically improve the versatility of the camera. For example, individual pixels can be programmed to adjust their light sensitivity in response to the amount of light available, thus avoiding saturation and increasing the dynamic range from darkest to lightest pixel.

Researchers are also looking at ways to provide on-chip image processing, such as edge detection, which would be useful for applications such as robotics. In work performed at ENSTA/LEI (Paris, France), Thierry M. Bernard is developing a programmable artificial retina (PAR) based on CMOS digital cameras.⁴ By processing the information within each pixel, such a PAR can react much faster in real time than devices with central processing. In this specific design, data propagates horizontally from one pixel to another, so that edges can be detected by rapid changes in flux and intermediate objects, such as image regions, can be passed along for higher processing.

One problem still requiring work is that of reducing the pixel-to-pixel variability in gain for CMOS-based cameras. The fixed-pattern noise that results from normal variation in threshold voltage, carrier mobility, oxide thickness, and gate length in the transistors used to amplify the signal is of little consequence for digital processing. For analog applications, including digital cameras, however, fixed-pattern noise can induce visible noise patterns.

There are several approaches to minimizing such noise sources. A unity gain amplifier can provide the required correction, but the device involves at least six field effect transistors per pixel. Another alternative is the active column sensor, in which the amplification and correction occur on a per-column basis, rather than a per-pixel basis.⁵

From camera to paper

While the electronic end of digital cameras has rapidly advanced, to fully compete with film cameras, digital cameras must offer the same level of convenience and quality in producing the final product, which for most people is still a print on paper. Here, the digital-camera support infrastructure is still being built.

One approach is to follow in the footsteps of the Polaroid instant print cameras and use essentially the same technology for quick digital prints. In July 2000, Polaroid and Olympus jointly announced the C-211 printing digital camera, which prints directly onto a Polaroid film cartridge, as well as receiving the scene digitally. Unfortunately for Polaroid, this digital camera, like its conventional predecessors, was designed for quick snapshots and official recording of scenes, not for high-quality art photography.

One complication comes from the fact that digital cameras, unlike color film, do not record all three primary colors in the same spot. Instead, color pixels are set side by side and must be combined to produce the final color image. Resolution inevitably decreases in the process, as does color fidelity.

Another problem comes from the generally limited memory storage of most digital cameras. Unlike film, which can be purchased interchangeably almost everywhere, memory storage units are customized for individual cameras.

Once into a computer system, photos can be printed out on a color printer, an advantage over conventional film. An alternative is to send the digital images to commercial printers who return finished prints, as with conventional film developing services.

So digital cameras will continue to make inroads into the consumer and professional photography markets.

REFERENCES

1. S. Ohno, et al., J. Imaging Sci. and Tech., 44, 51 (Jan/Feb 2000).
2. K. Mitani et al., SPIE Proc. 3965, 198 (July 2000).
3. B. Starr et al., SPIE Proc. 3965, 58 (July 2000).
4. T. M. Bernard, SPIE Proc. 3965, 277 (July 2000).
5. T. L. Vogelsong et al., SPIE Proc. 3965, 102 (July 2000)