• 
• 

Tuesday, July 28, 2009

Noise: Lose It, Part I

Eliminate noise at the source

This Article Features Photo Zoom

© 2009 John Paul Caponigro 9) Reduced signal/some floor noise

© 2009 John Paul Caponigro 10) Less signal/more floor noise

High temperatures exacerbate noise. Thermal energy (leakage current) in semiconductors can generate an electrical signal that’s difficult to distinguish from the optical signal. Ambient temperature increases leakage current by a factor of two for every 8 degrees Centigrade.

Underexposure results in more visible noise. Darker regions contain more noise than lighter regions in digital capture; the opposite is true of film. This is because the darker regions are recorded with fewer photons and less bits of data. It stands to reason that darker regions of images are produced with fewer photons. But why are they recorded with less bits of data?

Digital cameras record data in a linear progression. If a digital sensor is capable of recording 14 bits of data (or 16,384 shades of gray) with a dynamic range of 8 stops, the lightest stop contains half the data in the file; the next lightest stop contains half as much data; and so on. It generates this progression from dark to light—32 / 64 / 128 / 512 / 1,024 / 2,048 / 4,096 / 8,192. This means you want to expose to the right. Avoid the lower 2 stops whenever possible; they contain less than 1% of the total data in the file. With so little data, the signal becomes confused with the noise; the SNR is very low. What’s more, when these tonal regions are adjusted, brighter or contrastier, the noise contained there quickly becomes more pronounced, and with so few bits of data, it also has a tendency to posterize.

© 2009 John Paul Caponigro 11) Less signal/even more floor noise

© 2009 John Paul Caponigro 12) No signal/max floor noise

You’re better off making a light exposure and darkening the RAW file during postprocessing; this way, your shadows will be represented with much more data and contain less noise. How much should you expose to the right? A lot. There’s no perfect guide. Monitoring camera histograms is a useful practice, but it has limits. Why? The camera’s histogram is generated from a JPEG made on the fly. (If your camera allows it, set JPEG to low contrast for a more useful preview.) The RAW file contains more data than the JPEG and can be processed differently. Highlights can be aggressively mapped darker or recovered. The data is there. Lots of it. What you want to avoid is actually clipping highlight data; then there’s no data to recover. The histogram (and the “blinkies”) gives you an overly conservative approximation of when high-lights are clipped. Currently, no preview gives you a precise indicator of when highlight values are truly clipped.

The noise floor dominates other forms of noise. It’s created by the type of read circuits in the sensor, the transistor characteristics and support circuits such as the analog-to-digital converter. As light levels increase, the noise associated with light (“photon shot noise”) exceeds the noise floor. If the signal is increased by a factor of two (one ƒ-stop), the noise increases by a factor of one and the SNR increases by one. A higher SNR makes noise less visible. When the signal exceeds the maximum value the sensor is capable of capturing (dynamic range is a measure of the largest ratio of the capture signal to the noise floor), the noise drops because the signal is pinned at the saturation value.

John Paul Caponigro, author of Adobe Photoshop Master Class and the DVD series R/Evolution, is an internationally renowned fine artist and authority on digital printing. Get over 100 free downloads and his free e-newsletter  Insights at www.johnpaulcaponigro.com.