Automatic noise quantification for confocal fluorescence microscopy images

Due to photo-toxicity, fluorescence microscopy imaging is a trade-off between signal to noise ratio, total observation time and spatio-temporal resolution. We propose a new and simple method to quantify the signal-dependent and the non-signal-dependent components of the noise from a single fluorescence microscopy image. No reference image is required and the computation time allows on line quantification of the noise. The estimation is realized in two steps. We first estimate the signal-dependent noise by fitting the intensity of an estimated noise free image, computed by median filtering, to the estimated global noise variance. The second step estimates the signal-independent noise as the background variance, by computing the variance of the most homogeneous sub blocks of the image. After having shown the validity of our approach, we then use this method to quantify the noise in one experiment and show its correlation with the qualitative expected image quality.