ISO Sensitivity / Noise levels

ISO equivalence on a digital camera is the ability to increase the sensitivity of the sensor to enable faster shutter speeds and/or better performance in low light. The way this works in a digital camera is by "turning up the volume" on the sensor's signal amplifiers. Nothing is without its price however and doing so also amplifies noise and can also affect color saturation and hue.

The Sigma SD9 has three selectable sensitivities of ISO 100, 200 and 400.

Our noise comparison test involves shooting a colour patch chart (a GretagMacBeth ColorChecker) at the full range of ISO sensitivities and then measuring luminance and RGB noise at a 'mid' grey patch.

Note that this is a modification to our previous noise test which only measured luminance noise. The new test now provides a graph of luminance noise for each selectable sensitivity as well as individual RGB channel noise.

Sigma SD9 vs. Canon EOS-D60

Camera settings:

• Sigma SD9: manual exposure, High resolution, Shade WB, Sigma 50 mm F2.8 lens
• Canon EOS-D60: manual exposure, Default params, Cloudy WB, RAW, Sigma 50 mm F2.8 lens

Raw converter settings:

• Sigma Photo Pro v1.0: default processing, output 8-bit TIFF (sRGB)
• Canon File Viewer Utility: default settings, output 8-bit TIFF (sRGB)

Measurements taken at approximately 21°C (~70°F). Lighting was daylight. Crops below taken directly from TIFF files.

	ISO 100	ISO 100
	Sigma SD9 1/6 sec, F5.6	Canon EOS-D60 1/5 sec, F5.6
Original crop
Red channel
Green channel
Blue channel
	ISO 200	ISO 200
	Sigma SD9 1/10 sec, F5.6	Canon EOS-D60 1/10 sec, F5.6
Original crop
Red channel
Green channel
Blue channel
	ISO 400	ISO 400
	Sigma SD9 1/20 sec, F5.6	Canon EOS-D60 1/20 sec, F5.6
Original crop
Red channel
Green channel
Blue channel

As you can see noise levels are really quite similar, the most notable difference is that the SD9's noise is spread fairly equally throughout the color channels (as we would expect) where as the EOS-D60's noise is more visible in the red channel. The other thing which isn't immediately visible is the fact that the SD9's noise has a finer 'grain' to it, that's because each individual pixel can be affected by noise without necessarily affecting its neighbour, unlike with a mosaic sensor.

Luminance noise graph

Note that ISO sensitivity is on the horizontal axis of this graph. As noted above noise levels are fairly close and follow roughly the same line of increment.

RGB noise graph

Note that ISO sensitivity is on the vertical axis of this graph. As we noted above the SD9's noise is fairly equally balanced between each color channel, unlike a mosaic sensor which tends to have noisier red or blue channels.

Red hue noise

UPDATE May 2003: Today Sigma / Foveon posted a new version of the PhotoPro RAW conversion software, this appears to have vastly improved noise reduction / image output algorithms which have significantly reduced any specific noise in red hues. You can read my comparison of the old and new PhotoPro software by clicking here.