Interview with Frédéric Guichard of DXO Labs

09 March 2010

Q: What are the main challenges in keeping sharpness and good colour?

Firstly, sharpness and colour vary within the field, depending upon manufacturing variability and scene content and lighting. This variation was not visible for large pixels, but as pixel size has shrunk, more and more artefacts have become visible primarily for the following reasons:

• Manufacturing mechanical tolerances (centring of optical elements, surface errors on lenses, tilt, focusing error, sensitivity to temperature) do not scale, so their relative impact increases.
• Diffraction effect becomes significant.
• Camera module becomes thinner and is leading to higher CRAs.

Smaller pixel size is why keeping sharpness and good colour is increasingly a challenge:

• At 2.8μm, no correction was needed
• At 2.2μm, per-type calibration was needed
• At 1.75μm, per-unit calibration was needed to compensate for manufacturing variations
• At 1.4μm, the scene content needs to be taken into account

Secondly, low-power bulbs are now mandatory by law in Europe. This means that tungsten lamps are being replaced by dozens of different florescent lamps, each with a different spectrum. Per-unit calibration of colour lens shading is therefore no longer possible. Finally, scene variations (for example, light spectrum, scene reflectance, colour saturation, flare, and subject distance) have a visible impact on sharpness and colour. Automatic calibration-free correction is the way to go to solve these three challenges.

Q: Calibration is currently done in the production line - what are the limitations to this
approach?

Calibration in the production line has a significant impact on cost and on manufacturing. In particular, it is not used on 1/5'' (3Mp 1.4) and lower resolution because of cost, so the end-user has to live with the visible impact. Calibration in the production line is limited to a given number of light sources. It cannot cope with fluorescent light spectrum variation, so image quality depends upon the type of lamp used. Furthermore, calibration in the production line does not take into account scene variations: it
can accurately correct a target, but the quality of real-life shots varies depending upon the scene content.

Q: What factors do you need to consider in developing on-the-fly calibration?

On-the-fly calibration is constrained by silicon cost and response time. In particular, video requires continuous estimation while streaming, which means that statistics must be compact and analyzed in real time. Embedded implementation is therefore the key.

Q: How does this approach compare with production-line calibration in terms of cost?

Calibration-free correction eliminates or lowers production costs while keeping the silicon cost stable.

Q: What is the load in terms of processor use and power consumption for continuous
on-the-fly calibration?

Continuous on-the-fly calibration is implemented in hardware and does not use the processor. The hardware implementation is very compact and therefore its power consumption is negligible.

Q: Are the only applications in camera phones or are there other end-uses where this
approach would be advantageous?

Digital Still Cameras (most compact ones in particular) are starting to have issues with colour lens shading. However, DSCs have very complex zoom lenses, which would require too many shots and parameters to store to make per-unit calibration possible. So, DSC users have to live with colour lens shading until an approach such as on-the-fly calibration is adopted.

To find out more about the event, visit
http://www.image-sensors.com