row noise calibration

[wangchx67]

Hi, I want to query about the details of row noise calibration. Based on the description of paper, the first step is to extract the mean value of each row from raw data, and then sigma_r if obtained by maximize the log-likelihood. I am curious about which row mean value the maximizing log-likelihood of sigma_r is based on :

1. estimate every sigma_r of every rows and calculate the mean of all sigma_r
2. estimate single sigma_r based on the mean value of every row mean value
3. other

[Vandermode]

Hi,

estimate single sigma_r based on the mean value of every row mean value

this should be the correct choice FYI

[wangchx67]

Thanks！

[wangchx67]

Sorry to bother you again. Does "estimate single sigma_r based on the mean value of every row mean value" equals to "estimate single sigma_r based on the mean value of the whole image", cause the mean value of every row mean value equals to mean value of the whole image. Am I right?

[Vandermode]

“estimate single sigma_r based on the mean value of every row mean value" should be ”estimate single sigma_r based on every row mean values.“ I don't know why you call another mean here.

[wangchx67]

So the final sigma_r is obtained by the mean of all sigma_r estimated by the every row mean values?

[wangchx67]

For example, the var of row1 is sigma_r1, that of row2 is sigma_r1 ..., all_sigma = [sigma_r1, sigma_r1, ..., sigma_rh], and the calibrated sigma_r equals to mean(all_sigma)?

[Vandermode]

row mean values serve as the underlying measured data for fitting a statistical model. we used a simple maximum likelihood estimation to fit a Gaussian distribution from these data. mu and sigma can therefore be estimated with the Eq. (2) here