Short-Pass, Long-Pass, Bandpass, Neutral Density: A Practical Filter Guide for Machine Vision

The 30-Second Version: Four filter families solve 90% of machine vision filtering problems. Here's what each one does and when to reach for it.

The Language of Filters (Quick Vocabulary)

Before we talk filter types, let's get the terminology straight—because "green filter" can mean wildly different things depending on who's talking.

Transmission is simply how much light gets through, expressed as a percentage. This is your light budget.

Cut-on wavelength is where a long-pass filter starts passing light—the edge of its passband on the short-wavelength side. Cut-off wavelength is the opposite: where a short-pass filter stops passing light.

Half-power point (HPP) is where transmission drops to 50% of the peak value. This is the industry-standard way to define filter edges, and it's more precise than eyeballing where the curve "starts to drop."

Center wavelength for a bandpass filter is the midpoint of the passband, calculated from the two half-power points. It's not just a simple average—it's essentially a harmonic-mean style center. This matters because two filters both labeled "green" might behave very differently at the edges.

Bandwidth (often called FWHM, for "full width at half maximum") is the width of the passband between those half-power points. This tells you how picky your filter is.

One more note: the industry commonly specifies cut-on and cut-off locations at 5% absolute transmission. So "where it starts passing" is a defined criterion, not a vibe.

Long-Pass Filters

A long-pass filter does exactly what the name suggests: it passes wavelengths longer than its cut-on point and blocks shorter wavelengths.

The classic use case is IR illumination. You want your infrared LEDs to illuminate the scene while blocking visible ambient light from reaching the sensor. A long-pass filter set to cut on around 700–850 nm (depending on your IR source) lets your illumination through while rejecting the factory's fluorescent lights, sunlight through windows, and other visible interference.

Think of it as a bouncer who only lets tall people into the club. Everything above the height requirement gets in; everything below gets turned away.

Long-pass filters are also useful when you're working with near-IR or shortwave IR wavelengths and need to reject the entire visible spectrum. If your sensor sees wavelengths you don't want it to see, a long-pass filter is often the simplest solution.

Short-Pass Filters (Including IR-Cut)

A short-pass filter is the mirror image: it passes wavelengths shorter than its cut-off point and blocks longer wavelengths.

The most common machine vision application is the IR-cut filter for daylight color fidelity. Here's the problem it solves: silicon sensors are quite happy to detect near-infrared light that your eyes can't see. Without an IR-cut filter, your "color" camera captures wavelengths that mess up color accuracy—reds look wrong, skin tones go strange, and the image doesn't match what a human would see.

A short-pass filter (typically cutting off around 700 nm) blocks the near-IR while passing all the visible wavelengths. Your color image now represents what's actually there.

Think of it as a bouncer who only lets short people in. Everything below the cutoff passes; everything above gets rejected.

Bandpass Filters

A bandpass filter passes only a selected band of wavelengths and rejects everything outside it—both shorter and longer wavelengths.

This is your go-to when you want to isolate a specific LED wavelength from broad ambient light. Say you're using a 470 nm blue LED for illumination. A bandpass filter centered at 470 nm with a 20 nm or 40 nm bandwidth passes your LED light while rejecting the warm glow of overhead incandescent lights, the greenish cast of fluorescents, and the sunlight coming through the windows.

The narrower the bandwidth, the more precisely you isolate your wavelength—but remember the tradeoff from our interference filter discussion: narrower bands mean less transmission, even for the wavelengths you want.

Bandpass filters are also essential for fluorescence imaging, where you need to capture the emission wavelength while blocking the excitation wavelength. And they're useful for multispectral imaging applications where you're capturing images at several specific wavelengths to detect material properties.

Think of a bandpass filter as a bouncer with a very specific guest list. Only people within a certain height range get in; everyone else—too tall or too short—gets turned away.

Neutral Density (ND) Filters

A neutral density filter is spectrally boring on purpose. Its job is to reduce light intensity without favoring any particular wavelength—it attenuates everything equally across the spectrum.

Why would you want this? Sometimes your scene is simply too bright for your camera's dynamic range. Rather than stopping down the aperture (which increases depth of field and might not be what you want optically) or reducing exposure time (which might introduce motion artifacts), you can add an ND filter to reduce the light reaching the sensor.

ND filters are characterized by optical density, defined as D = log(1/T), where T is transmission. An ND filter with D = 1 transmits 10% of light. D = 2 transmits 1%. D = 3 transmits 0.1%. They're also sometimes specified by a filter factor used for exposure compensation.

Here's the beautiful thing about ND filters: the math is clean. When you stack filters, filter factors multiply and optical densities add. An ND 1 stacked with an ND 2 gives you ND 3. This means you can design attenuation like LEGO blocks instead of guesswork—buy a set of ND filters and combine them to hit whatever attenuation you need.

Putting It Together

When you need to block visible light and use IR illumination, reach for a long-pass filter. When you need accurate color and want to block IR, reach for a short-pass (IR-cut) filter. When you need to isolate your specific LED wavelength from ambient light, reach for a bandpass filter. When you need to reduce light intensity without changing color balance, reach for a neutral density filter.

And when you need to do several of these things at once? Stack them. The physics is on your side—filter factors multiply, optical densities add, and you can build the spectral response you need from standard components.

The key is knowing which family solves which problem, and understanding the tradeoffs each one brings.

This post is part of KUPO's technical education series on optical filters for machine vision. Questions about filter selection for your application? Contact our optical engineering team.