6. Noise filtering

6.1 Anscombe transform

Let's focus on the case of an image whose noise is of quantum origin (electronic noise is neglected, which is legitimate most of the time): if N represents the signal per pixel (in number of primary quanta absorbed), the average signal, then √ represents the noise level and (consequently) the signal-to-noise ratio. The Anscombe transform involves substituting its square root (N → 2√(N + 3/8)) for the signal. Francis Anscombe showed in 1948