Check for saturated pixels in ImageJ: histogram, HiLo LUT and the numbers

To check for saturated pixels in ImageJ, run Analyze › Histogram (Ctrl H) and compare Max with the highest value your camera can record: 255 for 8-bit, 4095 for a 12-bit camera, 65535 for a full 16-bit one. If Max sits exactly at that limit, and especially if Mode equals Max, pixels are clipped. To see where, apply Image › Lookup Tables › HiLo with the display range set to the full range: saturated pixels show red, pixels at zero blue.

Below, a sample image and a deliberately saturated copy of it show what each check looks like, and how much clipping changes the measurements.

What is a saturated pixel?

A pixel is saturated (or clipped) when the light reaching it was more than the camera could count: it records the maximum value, and everything brighter is lost. A saturated region is a flat plateau at the top value. Its true shape and brightness are unknown, and no software can recover them.

The limit depends on the camera, not only the file:

So a 16-bit file whose maximum is 4095 is not "far below 65535": it is probably a 12-bit camera at full scale. Check the camera's bit depth in its data sheet or in the file's metadata (Image › Show Info…).

A worked example: one image, and a saturated copy

The test image is a synthetic nuclei sample (16-bit, 512 × 512 pixels, 70 nuclei). Its brightest pixel is 6232, far below anything a camera would clip, so it is a fair "good" image. To make a "bad" one we ran Process › Math › Multiply… with a value of 15 on a copy: in a 16-bit image ImageJ clips every result above 65535 to 65535, which is exactly what an over-long exposure does. Without the ceiling the brightest pixel would have been 93,480.

You can do the same with any of your own images to see what saturation looks like in your measurements; just never measure the copy.

How do I read the ImageJ histogram?

Analyze › Histogram (Ctrl H) works on the active image or on the selection, if there is one. It draws the histogram, the count of pixels at each grey value, and lists:

The buttons below it: List gives the counts per bin as a table, Copy puts them on the clipboard, Log adds a log-scaled version on top (small counts become visible), and Live updates the histogram as you move a selection.

On a 16-bit image the histogram is binned between the image's minimum and maximum (316 to 6232 here), so the x axis is not 0 to 65535 and a spike at the top can hide inside the last bar. Read Max and Mode, not only the picture.

What does a saturated histogram look like?

The same window on the saturated copy:

Two numbers give it away: Max is exactly the ceiling, and Mode equals Max, because the clipped pixels pile up in the last bin. In a healthy image the mode is in the background, never at the top. The exact count at 65535 is 8,359 pixels, 3.19% of the image (the last bin, 8,442 pixels, also holds a few values just below).

On a linear plot the spike is easy to miss. Plotted on a log scale against what the scene really held, it is obvious: the histogram is cut off at 65535 and everything above is stacked into one bar.

To count the clipped pixels exactly in a macro:

getHistogram(values, counts, 65536);   // 16-bit: one bin per value (use 256 for 8-bit)
print(counts[65535]);                  // pixels at the 16-bit maximum; counts[4095] for a 12-bit camera

How do I see where the saturated pixels are? (the HiLo LUT)

HiLo is a grey lookup table with one change: the lowest value of the display range is drawn blue and the highest red. It shows where the extremes are without changing a pixel.

1. Image › Lookup Tables › HiLo. HiLo is not built into ij.jar: Fiji includes it, and plain ImageJ 1.x lists it only when a HiLo.lut file is in its luts folder (tables in that folder appear below the built-in ones). It is easy to make: grey, with entry 0 blue and entry 255 red.
2. Image › Adjust › Brightness/Contrast… (Ctrl Shift C), click Set and type the full range: 0 and 255 for 8-bit, 0 and 65535 for 16-bit (or 0 and 4095 for a 12-bit camera).
3. Red pixels are now at the maximum (saturated), blue pixels at 0 (clipped at the bottom, usually from too low an offset or an over-subtracted background).

The right-hand panel is the trap. Process › Enhance Contrast with 0.35% saturated (and, in a similar way, Auto in Brightness/Contrast) narrows the display range so that a small share of the pixels lie beyond its ends (here 460 to 5734), and HiLo then paints those red. They are only above the display maximum, not saturated. HiLo answers "is it saturated?" only when the display range is the full range.

How do I check Min and Max with Analyze › Measure?

For a number per image or per object, add Min & max gray value in Analyze › Set Measurements…, then press M (Analyze › Measure). With a selection, the numbers are for the selection only, so you can check each cell or band you measure.

With the ROI Manager, select all ROIs and click Measure: any row whose Max equals the ceiling is an object with clipped pixels. In our saturated copy, 37 of the 69 objects found by an Otsu threshold contained at least one pixel at 65535.

Why does saturation ruin intensity measurements?

Every intensity measurement assumes the pixel value is proportional to the light. A clipped pixel breaks that for the brightest objects only, so the error is not a constant you can correct:

A condition that doubles the signal of already-bright cells looks like a much smaller effect, and the brightest cells, often the ones you care about, are hit hardest. The same applies to western blot bands (a saturated band's integrated density is too low), to ratios and FRET, and to colocalisation, where clipped pixels distort Pearson's r and the Costes thresholds.

Thresholds suffer too: the pile-up in the last bin is a peak that histogram-based methods (Otsu, IsoData and the others, see threshold methods compared) take into account.

What to do about it

• At the microscope: use the camera software's own over-exposure display (most have one) or HiLo on a live image, and lower the exposure, gain or laser power until the brightest real structure is below the limit, with some margin. Keep these settings for every image you will compare.
• After acquisition: you cannot recover the data. Leave out objects with clipped pixels (and say so), or re-image.
• For figures only: some saturation for visibility is acceptable if no intensity is compared, and the legend says how the contrast was set.

Pitfalls

• Reading the picture, not the numbers. A 16-bit histogram is binned, and a spike can hide in the last bar. Read Max and Mode.
• Comparing Max with the wrong ceiling. A 12-bit camera saturates at 4095 inside a 16-bit file.
• HiLo after an automatic contrast stretch. Red then means "above the display maximum". Set the full range first.
• Clipped at the bottom. A Min of 0 with many pixels there (blue in HiLo) means the offset was too low or the background was over-subtracted; low values are lost too.
• Checking the whole image only. A field can be fine overall while the one cell you measure is clipped. Measure Max inside each ROI.
• Applying contrast. Brightness/Contrast › Apply rewrites the pixels and creates saturation that the camera never had.
• Processed files. A JPEG, a screenshot or an image exported for a slide has been rescaled; its maximum says nothing about the camera.

Reporting it

Images were acquired with [exposure, gain] set so that no pixel in the structures of interest reached the [12-bit] camera maximum; saturation was checked with the ImageJ histogram (Analyze › Histogram, maximum below [4095]) [and objects with clipped pixels were excluded from intensity measurements].

Journals increasingly ask for this; it costs one sentence. For a macro that checks thousands of files, or for reading the bit depth from a maker's own format through Bio-Formats, ImageJ and Fiji are the tools.

An easier way: check saturation in LabCanvas

LabCanvas Image Analysis checks every image for clipping the moment it opens, in the browser, with nothing to install and no LUT to load.

1. Open your image (TIFF, including 16-bit stacks).
2. In the Image tab, read Image check › Clipping: the share of pixels at the top value or at 0, per channel and inside the selected ROIs.
3. Click Show where to mark the clipped pixels in red and blue, like HiLo but without changing the display range; Check every slice goes through a whole stack.

That is the built-in bead Z-stack sample: nothing clipped, brightest value 3,688. The Display section has a histogram per channel with minimum and maximum sliders, and Auto saturates 0.35% of the pixels at each end of the display only; every measurement reads the raw values.

Why it is a good choice for checking images:

• It is automatic: no histogram to open, no Max to compare by hand. It knows the limits in the table above (255 for 8-bit; 65535, or the 10-, 12- and 14-bit camera maxima, for 16-bit), flags a pile-up at the maximum of any image, and warns once more than 0.1% of the pixels are clipped.
• More than saturation: the same panel checks the Pixels (bit depth, gaps in the histogram, JPEG or a picture from a figure), Illumination, Focus and Scale before you measure.
• Nothing to install: any computer or tablet with a modern browser.
• A methods sentence: the Methods line under each results table states the bit depth measured. It is free with a LabCanvas account, and every measurement is validated in more than 4,000 automated tests.

Open Image Analysis. ImageJ and Fiji are independent open-source projects; LabCanvas is not affiliated with them.

References

• Schneider CA, Rasband WS, Eliceiri KW (2012). NIH Image to ImageJ: 25 years of image analysis. Nature Methods 9, 671–675. doi:10.1038/nmeth.2089