2. Noise Variance Stabilisation

Introduction

SReD's workflow begins with the application of the Generalised Anscombe Transform (GAT) to stabilise noise variance Boulanger et al., 2010. This step addresses the noise in microscopy images, which generally exhibit a combination of Poisson and Gaussian noise. The GAT nonlinearly remaps pixel values to produce an image with near-Gaussian noise and stabilised variance, preserving local contrast and overall image statistics.

This stabilisation is essential for robust downstream processing, mitigating violations of normality, homoscedasticity, and outlier assumptions that can compromise correlation metrics. For more information on how the GAT works, please refer to Boulanger et al., 2010 and the SReD paper.

On-demand noise variance stabilisation

The GAT is applied automatically in SReD's functions but can also be applied independently by opening the input image and clicking on "Plugins>SReD>Noise variance stabilisation". A window will appear for the user to define some values. These are the initial guesses ("gain", "offset" and "noise standard deviation"), where an optimiser will begin its calculations. The default values work for most applications.

Select the plugin	Choose initial guesses

Alternatively, the user can pre-define a ROI containing only camera noise (i.e., a dark region), and by checking the "Estimate offset and StdDev from ROI?" checkbox, SReD will attempt to retrieve some parameter values from the data itself.

Alternatively, use a ROI to estimate initial guesses from the image

Once the function is finished calculating, the output will appear automatically. It consists of a variance-stabilised image where the contrast is kept but the data's range is modified so that the noise variance is as close as possible to 1.

Once SReD is finished, the variance-stabilised image appears automatically

NOTE: SReD automatically stabilises the noise variance when its functions are used. The ability to perform noise variance stabilisation independently is a feature that allow users to pinpoint the origin of potential problems in subsequent analyses, since the GAT can sometimes behave unexpectedly.