Open griffinfoster opened 9 years ago
parameter estimation for general Gaussians seems to be a bit of a pain. On the other hand, the Poisson distribution looks a bit more promising. The Cauchy and Gaussian distributions are pretty straightforward; none has have skewness and the kurtosis is easy to compute.
Poisson and Gaussian have easy to compute skew and kurtosis once you have fit for the parameters.
Or, one can compute the 'skewness' and 'kurtosis' from the moments by making an assumption about the distribution, which we can at least start by approximating as normal.
On Thu, Apr 2, 2015 at 10:34 PM, Sphesihle Makhathini < notifications@github.com> wrote:
parameter estimation for general Gaussians seems to be a bit of a pain. On the other hand, the Poisson distribution looks a bit more promising. The Cauchy and Gaussian distributions are pretty straightforward; none has have skewness and the kurtosis is easy to compute.
— Reply to this email directly or view it on GitHub https://github.com/SpheMakh/Fidelity/issues/3#issuecomment-89038080.
An ideal residual image will have Gaussian noise (with some bias towards positive values be cause sources have positive flux). We can fit multiple (or general) distributions to the histogram of the residual image pixel values. Then ask which distribution best models the residual noise, and if that is close to Gaussian.
It would be useful to try general Gaussian, Lapacian, Cauchy distributions. And compute higher order moments such as skew and kurtosis.