Professor Newman on…
Introducing a new graph for visualising how noisy image files look and comparing different cameras
If you have the right equipment, physical measurements of technological articles are quite easy. Image sensors are particularly straightforward, because they are themselves measuring devices. So long as it is known how they should respond to a stimulus, how they actually do tells most that there is to know about performance. But that depends on what you think ‘performance’ means. The image sensor industry relies on several metrics to measure sensor performance. These include quantum efficiency, sensitivity and read noise. But the question is, how are these relevant to photographers?
Quantum efficiency measures the effectiveness of the sensor in turning incident light photons into measurable charge. This affects how noisy an image looks, so it’s relevant to photographers. But it’s not the sole determinant of noise, so taken in isolation it can be misleading.
Sensitivity (not to be confused with photographic ISO) relates the amount of incident light to the voltage output from the sensor. It’s useful to electronic designers of ancillary circuitry. Now that most sensors have a digital output it is mostly irrelevant, but it is still given.
Read noise is a measure of the electronic noise added to the signal derived from the light. This is relevant to photographers, but not as much as might be thought. Most noise in an image comes from the light itself, due to the fact that light is a particulate phenomenon and arrives randomly, in discrete packets of energy called photons. If there are not many photons to work with, their randomness is apparent in the image. This is why increasing the number of photons recorded (by increasing quantum efficiency, exposure or sensor area) makes an image look less noisy. In contrast, read noise is only really visible when the amount of light is small. In the end it provides the limit to acceptable noise in the deep shadows, but that’s all.
So, if photographers need not be interested in the industry measurements for sensors, what would be more appropri
