Flat-Field Correction

corrected = (raw − dark) ÷ (flat − dark) × mean(flat − dark)

The dark frameLoading... is subtracted first from both the image and the reference, because the camera's offset is not light and must not be divided. Dividing by the dark-subtracted flat then removes the illumination pattern: a pixel that receives 70% of the center's light in the flat is divided by 0.7. Multiplying by the mean of the flat restores the original units, so blank glass in the corrected image reads the same average brightness it did before.

The correction is applied per channel in a color image, which also removes small color differences between center and edge.

Use a clean, blank area of the slide, through the same objective and with the same lamp setting, filters, condenser position and exposure as the images it will correct. Defocus slightly: dust and debris on the slide blur into nothing, while the illumination, which is not in the specimen plane, is recorded unchanged.

Average several frames. Noise in the flat is divided into every corrected image, so a single noisy flat adds its noise to everything; averaging 16 frames reduces it by a factor of 4. Anything left in the flat, a speck, a fiber, tissue, appears in every corrected image as a bright ghost, so inspect it before use.

Correct a second, independent image of blank glass and measure what is left. The corner-to-center ratio should be within about 2% of 1.0, and the variation across the field should be small and structureless. In a tiled scan, a corrected blank or lightly stained area should show no tile-shaped grid.

A reference is valid only for the light path it was captured on. Capture a fresh one when you change objective, lamp setting, filter or condenser position, after realigning Köhler illuminationLoading..., and at least once per imaging day. In fluorescence the reference is a uniformly fluorescent slide instead of blank glass, and it also corrects pixel-to-pixel sensitivity differences in the sensor.