Fig 1.
Estimating the gain parameter, g, of the sCMOS camera: Variance vs mean plot of bright field images acquired during the Illuminated white wall experiment (see ‘Experiments’ section).
Intensity levels were incremented in 10% steps from 10% to 100%. The gain is calculated from the slope of plot using the Eq (19), yielding (shown as the solid cyan line). The y-intercept of the linear fit is 6.4. This value is consistent with the value 4.7 obtained using Eq (20).
Fig 2.
Estimation of read noise parameters, and
, for the sCMOS camera.
Panel (a) shows the Pearson correlation coefficient, Eq (24), between the empirical quantiles , , from the dark frame image (acquired during Cap-On experiment) and the unscaled theoretical quantiles, F, for a range of shape parameters (
). The dashed line in the figure represents
corresponding to the maximum PCC score. Panel (b) plots the dark image frame empirical quantiles against the theoretical unscaled quantiles at the optimal
from panel (a). The slope of this curve gives estimation of the scale parameter (
) given by
. The y-intercept of the fitted line is 0.009 (which is close to the expected value 0).
Fig 3.
(a) An sCMOS image of fluorescently labelled single DNA molecules deposited on a glass slide.
The image is acquired using a procedure described in the subsection Experiments in Materials and Methods. The image is split into tiles of size 64x64 pixels, where each tile is given a label {row,column}, where in this example row,column = 1,...,16. (b) A histogram of the image counts for a single tile, here tile {7,13} (yellow bordered tile in Fig 3(a)). The blue bars represent pixels regarded as true background, while the orange bars represent the outliers (not true background or signal pixels). The background Poisson parameter is estimated to . The image count threshold was estimated to be
. This threshold separates the blue and orange bars and was determined using a p-value threshold, pGoF = 0.01, for the goodness-of-fit tests. The dashed black curve shows the fitted PMF for the estimated background, extended to the full range of image counts (in our method, we fit a truncated PMF to the blue bars). (c) To show the contrast across the tiles in the image, we estimate
for another tile {8,8} (green bordered tile in Fig 3(a)) with
. Thresholded and segmented versions of the image from panel (a) are found in the Supporting information, S1 Fig.
Fig 4.
Performance of the sCMOS-PMF algorithm at low exposure time images: (a) 1 ms, (b) 8 ms, and (c) 20 ms (tile ).
The sample being imaged is identical to the one in Fig 3. Black dashed curves represent fitted PMFs. Estimated background Poisson parameters () and threshold counts (
) are given in the figure legends for each case.
Fig 5.
Relationship of the estimated background Poisson parameter () with the exposure time of the sCMOS camera.
We show the average background Poisson parameter (over all tiles) for images with exposure times (1–1000 ms). The inset shows zoomed version of the image for 1–16 ms. Notice that increases linearly with exposure time, as it should (since the number of collected photons increases linearly with exposure time). The error bars are the standard deviations of
across tiles in the image.
Fig 6.
The photophysical sCMOS image processing pipeline applied to bacteria cells overexpressing GFP.
(a) a 100 ms exposure time image of fluorescently stained cells (balanced gain setting). (b) Binarized image processed by our unsupervised thresholding algorithm with a p-value threshold of pbinarize = 0.01. Our algorithm is designed so that for this choice of threshold we expect approximately 1% false positives, which from visual inspection may roughly be the case (no ground truth is available here). (c) Output of our segmentation approach with the yellow pixels forms the boundary of the “objects” identified by our unsupervised segmentation method. Example images at lower and higher exposure times are found in the Supporting information, S2–S3 Figs.
Fig 7.
Comparison of the mean background Poisson parameter per pixel area () between sCMOS and EMCCD cameras.
Images were recorded by sCMOS and EMCCD cameras focusing on same field of view (see Dual camera same FOV experiment in experiment subsection). For each camera, was first calculated in individual image tiles; plotted values represent the mean over all tiles, with error bars showing the standard deviation of
across tiles. The Poisson parameter is scaled by the pixel area as the pixel size is different for the two cameras (see Experiments section). The calibration parameters for the sCMOS camera are same as calculated in the parameter estimation section while for the EMCCD camera the EM Gain knob on the camera was set at 100. The EMCCD calibration parameters , gain/AD factor (g/f), offset (Δ), read noise (r0) are 4.518 , 483.77, and 7.81 respectively, calculated following the procedure discussed for EMCCD cameras in [10].