Dark Frame

Three things appear in an image taken with no light. The offset, also called the bias or pedestal, is a fixed value added on purpose so that noise around zero is not clipped; on scientific CMOS cameras it is often set around 100 counts. Dark current is charge generated by heat, which grows with exposure time and temperature; cooled cameras keep it small. And fixed-pattern structure, pixel-to-pixel differences in offset, stays the same from frame to frame.

Averaging many dark frames gives each pixel's offset; the frame-to-frame spread of each pixel gives its read noiseLoading.... Neither is light, and both must be removed before any calculation that treats pixel values as proportional to light.

Use the same exposure timeLoading..., gain, readout mode and camera temperature as the images the dark frame will correct. Block all light: lamp off or shuttered, eyepiece shutter closed, and room lights low, since light can reach the camera through the eyepieces. Average several frames, at least 16 for a correction frame, and around 100 if you also want a read-noise measurement.

Re-capture when exposure, gain or temperature changes. A dark frame from a different exposure carries the wrong dark current, and one taken with stray light in it subtracts too much.

The offset matters little where the image is bright and a great deal where it is dark. Worked: with an offset of 100 counts, blank glass at 2,000 counts and a strongly stained pixel at 300, the true values are 1,900 and 200. The correct optical densityLoading... is log₁₀(1,900 ÷ 200) = 0.98. Without the dark subtraction it comes out log₁₀(2,000 ÷ 300) = 0.82, about 16% low.

The error grows with stain strength, so it compresses the top of the range, exactly where strong and very strong staining are told apart. In flat-field correctionLoading... it also biases the division, leaving the darkest corners under-corrected.