Figure 4. Global registration precision.... | Rockefeller University Press

Figure 4.

$Figure 4. Global registration precision. (A and B) Distribution of remaining distances between n = 7,387 registration features after registration with a rigid 2D transformation model for a single cell. No bias (mean value is zero and distribution symmetric) was introduced and the width σ for x (A) is 0.44 pixels or 70 nm and for y (B) is 0.39 pixels or 62 nm. A and B represent the error of the lateral alignment in x and y as laid out in Fig. 2 for a single cell with a target of σ = 0.5-pixel global registration precision. (C) Two kinds of beads were embedded in an agarose gel. First the diffraction-limited beads were aligned using a Gaussian fit for the center of the bead location, which was then used for alignment. Alternatively, the beads (1-µm diameter) were aligned using a quadratic fit for corresponding features. The remaining mean error after registration relative to the σ used for fitting is shown. The differences between both methods are within the noise level at <1 nm. The total error of the alignment is dominated by the localization precision (here 10 nm; error bar in C).$

Global registration precision. (A and B) Distribution of remaining distances between n = 7,387 registration features after registration with a rigid 2D transformation model for a single cell. No bias (mean value is zero and distribution symmetric) was introduced and the width σ for x (A) is 0.44 pixels or 70 nm and for y (B) is 0.39 pixels or 62 nm. A and B represent the error of the lateral alignment in x and y as laid out in Fig. 2 for a single cell with a target of σ = 0.5-pixel global registration precision. (C) Two kinds of beads were embedded in an agarose gel. First the diffraction-limited beads were aligned using a Gaussian fit for the center of the bead location, which was then used for alignment. Alternatively, the beads (1-µm diameter) were aligned using a quadratic fit for corresponding features. The remaining mean error after registration relative to the σ used for fitting is shown. The differences between both methods are within the noise level at <1 nm. The total error of the alignment is dominated by the localization precision (here 10 nm; error bar in C).