Clipping | Imaging Glossary

View

Definition

A pixel is clipped (or saturated) when the light reaching it would produce a value above the largest number the camera can record — 255 in an 8-bit image, 4,095 at 12 bits, 65,535 at 16 (see bit depthLoading...). Every brighter level is recorded as that same maximum, so the differences between them are lost and cannot be recovered afterward.

In a color image, clipping can hit one channel before the others. The ratio between red, green and blue is then wrong, which corrupts white balanceLoading... and any optical densityLoading... or color deconvolutionLoading... measurement in that area. Check the histogramLoading...: blank glass should sit high but clearly below the maximum in every channel. Reduce exposureLoading..., use a neutral density filterLoading..., or lower gain — not lamp voltage.

Learn by playing

Larry's White Balance Lab

Game · White balance · Color temperature · Neutral density filters

Max at 255

Game · Exposure and clipping · Bit depth · DAB non-linearity

Bring Your Own Slide

Game · Image quality check · White balance · Flat-field correction

Two ceilings, and either one loses data

A pixel can hit a limit in two places. The sensor's full wellLoading... is the most charge a pixel can hold; beyond it, extra light is not recorded. The analog-to-digital converter has its own ceiling, set by the bit depthLoading...: 255 at 8 bits, 4,095 at 12, 65,535 at 16. Cameras are usually set up so one of these is reached first, and either way the result is the same: every brighter level is recorded as the same number.

Because those levels are merged, nothing after acquisition can separate them again. Contrast adjustment, flat-field correction and color deconvolution all operate on the stored numbers, and a clipped pixel has already lost the information they would need.

Simplified

Every camera has a brightest value it can record, 255 in a standard 8-bit image. Anything brighter than that is written as 255 too. A pixel that was a bit too bright and one that was far too bright end up looking identical.

Once that has happened, no software can tell them apart again. The detail is gone at the moment the picture is taken.

One channel can clip before the others

In a color image, the three channels reach their ceilings at different light levels. Under a halogen lamp with heavy blue gain, or after white balanceLoading... has pushed one channel up, a single channel can clip while the other two still have room. The pixel's color is then wrong, because the ratio between channels is wrong.

That matters doubly for brightfield quantification. If blank glass clips, the software's estimate of its true brightness is too low, and every optical densityLoading... computed from it is shifted. If the clipped channel is the one color deconvolutionLoading... relies on, stain estimates in those pixels are wrong in a way no later correction can repair.

Simplified

Red, green and blue each have their own ceiling, and one can hit it before the others. When that happens the color of the pixel changes, because one color stopped growing while the other two kept going.

For stain measurement that is a real problem. If empty glass is clipped, the software gets the wrong idea of what pure white is, and every stain number calculated from it is off.

How to check before you acquire

Look at the histogramLoading... of blank glass in every channel, not only the combined image. Aim for the glass to sit high but clearly below the ceiling, around 80–90% of the maximum (roughly 200–230 in an 8-bit image), so normal variation from field to field never reaches it. Many acquisition programs can overlay saturated pixels in a warning color; turn it on.

If anything clips, reduce the exposure timeLoading..., add a neutral density filterLoading... or lower the gain. Do not lower the lamp voltage, which also changes its color. And avoid auto-exposure for quantitative work: it changes the exposure with the content of each field, which makes fields incomparable.

Simplified

Before you start, look at the brightness graph of empty glass for each color. The empty glass should be bright but not right at the top, around 80 to 90 percent of the maximum. Many programs can paint clipped pixels red so you can see them.

If something clips, shorten the exposure or add a gray filter. Do not dim the lamp, and do not use automatic exposure, which changes from field to field.

Explore TissueFAXS

Imaging Spectroscopy Platform

Precision whole-slide imaging with TissueFAXS, optimized optics, and multi-channel acquisition.

Two ceilings, and either one loses data

One channel can clip before the others

How to check before you acquire

Related Terms