Image filters

Filtering crosses physics, electronics and computation. An interference filter in an optical path passes a band of wavelengths and rejects the rest, separating the light you want from the light you don't. An electronic filter in a circuit does the same with a band of frequencies in the time domain. An image filter does the same job in the spatial domain, separating the spatial signal you want to keep from the spatial signal you don't.

In every case the principle is the same: the signal you want and the signal you don't rarely occupy the same place in the relevant domain. Real biology (the cells, the structures, the boundaries) lives at one set of spatial frequencies; sensor read noise lives at another, bright outliers from hot pixels or stray particles at a third, and slowly varying autofluorescence at a fourth. The right filter is the one whose pass-band matches the spatial signal you want to surface. That pass-band can be explicit (a Gauss filter passes low spatial frequencies) or implicit (a median filter passes neighborhoods that aren't dominated by outliers).

The unwanted signal isn't always uniform. It can be a steady floor across the field, a gradient that varies smoothly across the image, or a localized region of its own. Each case suggests a different filter, sometimes a chain of filters, sometimes position-aware filtering. Recognizing the character of that signal (its scale, its uniformity, whether it is random or structured) often determines the right filter more than the filter's name does.

The fastest path through the filter family starts from the structure you want to surface. Here are a few common cases.

Isolated extreme pixels alongside the signal you want: the Median filterLoading... sorts the neighborhood and picks the typical value, so it ignores outliers by construction. Choose it when individual pixels carry signal you don't need to keep.

Broad, low-amplitude variation across the field: the Gauss filterLoading... averages the neighborhood with a smooth weight profile. It reduces the variation and keeps the larger features that define the image.

Smoothing across regions while keeping edges crisp: the Bilateral filterLoading... weights neighbors by spatial distance and by intensity similarity, so smoothing happens within regions and stops at boundaries. Choose it when the transitions are part of the structure you care about.

Surfacing edges (cell boundaries, vasculature, structural transitions): use SobelLoading... (horizontal, vertical or cross) for directional derivatives, CannyLoading... for clean, thin edge lines, or LaplaceLoading... for second-derivative responses. Each highlights a different kind of intensity change.

Surfacing membranes specifically: the Membrane detectorLoading... is tuned to the rapid double-variation pattern that membranes produce.

Sharpening for emphasis: use the Sharp filterLoading..., which enhances detail by raising contrast. The Unsharp filterLoading... does a different job: it removes high background by subtracting a blurred copy of the image. Both change the original intensities, so consider their effect on quantitative analysis wherever those intensities are the measurement.

A specific kernel in mind: the Custom filterLoading... lets you write the weights yourself. It is the clearest path when no textbook entry fits.

The StrataQuest Engines Reference has the full per-engine detail; this page maps the territory.