Hi, PRNU does not play a role, because the noise calculation takes only temporal noise into account. Non-linearity plays a role (you can see that also in the mean-variance plot), but more important is the limited dynamic range. In the mean-standard deviation plot you can observe that the curve is getting steeper and steeper if you move to larger effective output signals. So the curve has the tendency to reach a slope of 0.5 but right before coming to this point, the pixels saturate. Do you need more info about this effect ? I do have images and data available of devices with a larger dynamic range, and in that case getting a slope of 0.5 is no issue.
What are the reasons for a deviation from a slope of 0.5 in standard deviation versus mean log-log plot? PRNU, non-linearities, others?
ReplyDeleteHi, PRNU does not play a role, because the noise calculation takes only temporal noise into account. Non-linearity plays a role (you can see that also in the mean-variance plot), but more important is the limited dynamic range. In the mean-standard deviation plot you can observe that the curve is getting steeper and steeper if you move to larger effective output signals. So the curve has the tendency to reach a slope of 0.5 but right before coming to this point, the pixels saturate.
ReplyDeleteDo you need more info about this effect ? I do have images and data available of devices with a larger dynamic range, and in that case getting a slope of 0.5 is no issue.