0

For an experiment, the standard deviation (SD) is entered as a parameter for generating gaussian noise. The SD value is mentioned to be 25 when data (a matrix) has integer values in [0,255]. If the data was scaled to be within [0,1], what would be the value of the SD parameter? Would it be [25/(max of data values)]?

VanguardExplorer
  • 13
  • I am confused. First you say that the standard deviation is entered as a parameter, and then you say it is measured. These cannot both be true. First you say that the noise is Gaussian, and then you say it is confined to [0, 255]. These cannot both be true. I will answer ignoring your first sentence. – apt1002 May 10 '14 at 19:01
  • I will try to clarify.

    We have an original (clean) data matrix A, with integer values in [0,255].

    The experiment is conducted by: First, generating a zero-mean gaussian noise matrix B. The Standard Deviation (SD) of this noise is 25.

    Second, the data that is input to a recovery algorithm is C=A+B.

    The experiment is about testing the recovery algorithm by increasing the SD of B until results become unsatisfactory.

    The core issue is that the software I use has built-in functions that expect input in [0,1]. I wasn't sure about how to change noise SD when data is scaled to [0,1].

     – VanguardExplorer May 10 '14 at 19:19
  • Thank you, that is a different question, and much clearer. My answer below still works, as it happens! – apt1002 May 10 '14 at 19:30

1 Answers1

0

To scale the data from $[0,255]$ to $[0,1]$, you use a scale factor of $1/255$. The standard deviation scales with the data, so the standard deviation of the scaled data will be $25/255$.

apt1002
  • 2,033