Is our noise measurement affected by temperature?

Written by:

I’ve read a lot of articles about image sensors over time. Many of them were talking about something called “dark current noise”. dark current is electrons that are captured by the sensor but not as a reaction to light. This is actually unwanted signal the sensor has picked up. This unwanted signal has a fluctuation, which means this signal is creating noise when looking at more then one pixel. Because this signal is very weak, noise signal is only building up to a noticeable level when the exposure time is very long. Further more, dark current level is said to be reduced substantially if the sensor is cooled down.
We are only interested with what happens when we take noise measurement in our labs for our noise analysis database. The temperature in our labs is monitored and kept at a certain level. We have an air condition and a thermometer, so we are pretty much covered, and we always try to bring room temperature to about 68 °F (20 °C) when we do noise measurements. The question is, how strictly do we need to follow the 68 °F rule? Does temperature in an exposure time of 1/30 makes any difference? Does dark current noise effect our noise readings?

In order to test this, we made a simple test. We took two noise measurements, one when the room temperature was 66.2 °F (19 °C) and a second test when heating the room temperature to 84.2 °F (29 °C).
For this task we used the Olympus E-330 DSLR, set to 1600 ISO with Noise reduction set to OFF. The E-330 is using a CMOS sensor, and a quite noisy one, if I may add.

The full procedure of the test is as followed:

1. Set the room temperature to 66.2 °F (19 °C)
2. Take 3 JPEG shoots of the GretagMacBeth ColorChecker 24 test chart, exposure time is set to 1/30 and the camera is set to 1600 ISO NR OFF
3. Set the room temperature to 84.2 °F (29 °C) and let the camera warm up for about 45 minutes.
4. Take the same 3 image as paragraph 2, same exposure time.
5. For each temperature, read std. from the 6 monochromatic zones of the ColorChecker chart and average every set of 3 shots out.
7. Plot the Luma values to a graph (shown below).


As you can see from the graph above, a 18 °F difference hasn’t affected the results much. There is a slight deference in zone 6, but not in the expected way. Zone 6 has a std. value of 2.62 in the temperature of 66.2 °F, whereas in temperature of 84.2 °F zone 6 has a value of 2.56 std. lower when warmer. This could be explained by the fact that electronic devices tend to work less efficiently in warmer conditions, so our sensor might be less sensitive when working in 84.2 °F. Because it is slightly less sensitive, there is going to be less photon noise (This is the main noise in the image).

The table below shows the full noise readings, as you can see there the std value is slightly lower in about every zone of the chart. So the theory above could be just true. In any way, we have determined that room temperature is not critical when measuring noise, and that dark current is nothing to worry about at an exposure time of 1/30.

ZONE / Temp. 66.2 °F (19 °C) 84.2 °F (29 °C)
ZONE 1 0.19 0.18
ZONE 2 0.29 0.29
ZONE 3 0.5 0.49
ZONE 4 0.83 0.82
ZONE 5 1.52 1.51
ZONE 6 2.61 2.56

Digg this story   Add to del.icio.us





Related Posts:


Comments are closed.