Signal Processing Stack Exchange is a question and answer site for practitioners of the art and science of signal, image and video processing. It only takes a minute to sign up.
Sign up to join this community
$$\hat f(x, y) = g(x, y)-\frac{\sigma_n^2}{\sigma_L^2}\left[g(x, y)-m_L\right]$$
What are the meanings of the following terms:
Here we see that $m_L$ is subtracted from the Image and then the whole term is multiplied by $\frac{{\sigma_\eta}^2}{{\sigma_L}^2} $. Then, the whole term is again subtracted from $g(x,y)$.
What does that actually mean?
A wild guess. $L$ denotes a region $R$ around pixel $g(x,y)$, potentially a symmetric one of size $(2L+1)\times (2L+1)$. $m_L$ and $\sigma_L^2$ are the average and variance in $R$. And $\sigma_n^2$? Probably the estimated noise variance over the whole image, supposed stationary.
In a flat region with noise $\sigma_L^2 \sim \sigma_n^2$, the result could be close to the average. Around edges, one expect an higher response, so sharper. Potential problems arise when $\sigma_L \ll \sigma_n $. Another interpretation:
$$\hat{f}(x,y)=(1-\frac{\sigma_n^2}{\sigma_L^2})g(x,y)+ \frac{\sigma_n^2}{\sigma_L^2}m_L(x,y)$$ is a sort of weighted mean (since weights sum to one) of the image and its smoothed version, whose dependency on $(x,y)$ is made more explicit.
