Glossary

These are the terms that come up most often in this site’s articles, grouped by topic. The link after each definition leads to an article where the term is used in an actual calculation.

Optics and Lenses · Light and Lighting · Cameras and Sensors · Decision Algorithms · Edge AI and Systems

Optics and Lenses

Working distance (WD)
The distance from the front of the lens to the surface being inspected. The space for lighting and fixtures is decided within this distance.
Related article: The Camera Cannot Be Bolted On Last — Reserving Space and WD for Vision Cameras and Lighting at the Mechanical Design Stage

Field of view (FOV)
The width and height of the area on the target captured in one image. With the same sensor, a wider field of view makes each pixel cover a longer distance.
Related article: How to Calculate Machine-Vision Lens Focal Length

Focal length
The distance at which parallel light passing through a lens converges to a point. With the same sensor, a shorter focal length sees a wider area.
Related article: How to Calculate Machine-Vision Lens Focal Length

Magnification
The size of the image on the sensor divided by the size of the object.
Related article: Machine Vision Lens Calculator — Catching the FOV Error the Shortcut Formula Hides, Plus Depth of Field and Diffraction in One Pass

Object-side pixel resolution (µm/px)
The length one pixel covers on the object, found by dividing the field of view by the number of horizontal pixels and written in µm/px. It is the basis for judging how many pixels a defect occupies.
Related article: How to Calculate Machine-Vision Lens Focal Length

Depth of field (DOF)
The range of depth that stays sharp in front of and behind the focus distance. Stopping down widens it but lets in less light.
Related article: The Tug-of-War Between Distortion Correction and Depth of Field: How Telecentric Lenses Catch Micro-Defects Without Losing Them

F-number
Focal length divided by the aperture diameter. Each stop down roughly halves the light coming in.
Related article: In SWIR, One Stop Down Erases a 3.4 μm Pixel

Circle of confusion
The diameter up to which a point blurred by defocus is still treated as sharp. It is the reference value for depth-of-field calculations.
Related article: Machine Vision Lens Calculator — Catching the FOV Error the Shortcut Formula Hides, Plus Depth of Field and Diffraction in One Pass

Airy disk
The small disk into which light spreads even through a flawless lens. Its diameter is about 2.44 × wavelength × F-number.
Related article: In SWIR, One Stop Down Erases a 3.4 μm Pixel

Diffraction limit
The ceiling on resolution that even an ideal, aberration-free lens can reach. It comes from light spreading at the edge of the aperture, so no lens design can exceed it.
Related article: In SWIR, One Stop Down Erases a 3.4 μm Pixel

MTF (modulation transfer function)
A curve showing how much of the light-dark difference of fine line patterns a lens preserves at each line spacing. As sensor pixels shrink, this curve sets the effective resolution.
Related article: In the Age of 8K Line-Scan Cameras, the Lens Still Sets the Ceiling on Inspection Resolution

Telecentric lens
A lens whose object-side chief rays run parallel to the optical axis, so the object barely changes size when it moves slightly forward or back. It is widely used for dimensional measurement.
Related article: The Tug-of-War Between Distortion Correction and Depth of Field: How Telecentric Lenses Catch Micro-Defects Without Losing Them

Distortion
How far image positions shift from their ideal places toward the edge of the frame, usually given in %.
Related article: The Tug-of-War Between Distortion Correction and Depth of Field: How Telecentric Lenses Catch Micro-Defects Without Losing Them

Image circle
The diameter of the circle over which a lens projects an even image. If it is smaller than the sensor diagonal, the four corners come out dark.
Related article: How to Calculate Machine-Vision Lens Focal Length

Liquid lens
A lens that moves its focus by changing the curvature of a liquid interface with an electrical signal, without moving parts.
Related article: Liquid Lens Autofocus: Solving the Focus Problem on High-Mix, Low-Volume Production Lines

Light and Lighting

Reflectance
The share of light hitting a surface that bounces back. Under the same lighting, a low-reflectance surface comes out darker.
Related article: Raising the Camera Leaves the Glint of Its Built-in Light in Place

Specular reflection
Reflection from a smooth surface that leaves in one direction at the same angle the light came in. If the camera sits in that direction, a blown-out white spot appears.
Related article: Raising the Camera Leaves the Glint of Its Built-in Light in Place

Diffuse reflection
Reflection in which light scatters in many directions from a rough surface. It is why a matte surface looks about equally bright from any angle.
Related article: When the Light Dims by 5 %, How Many Good Pixels Turn Into Defects?

Illuminance (lx)
The amount of light reaching a surface, measured in lx. From a point source, doubling the distance cuts it to a quarter.
Related article: Raising the Camera Leaves the Glint of Its Built-in Light in Place

Contrast
The difference in brightness between a defect and its background. If lighting and optics do not create it, the algorithm has nothing to find.
Related article: Machine-Vision Inspection: Physics Comes Before Software

Coaxial lighting
Lighting that uses a half mirror to shine along the same axis as the lens. It makes flat, mirror-like surfaces look evenly bright.
Related article: Is Coaxial Light Enough for Fine Scratches on Polished Metal? — Redesigning Coaxial Illumination Around Half-Mirror Loss and Tilt Tolerance

Dome lighting
Lighting that illuminates evenly from many directions with light reflected off the inside of a hemisphere. It reduces glare on curved surfaces but can leave a dark spot under the viewing hole.
Related article: Why Does a Black Spot Remain in the Center Under a Dome Light? — Solving Curved Specular Illumination with Viewing-Hole Geometry, Coaxial Fill Light, Dome Diameter and WD

Dark-field illumination
Lighting from a low angle that keeps flat areas dark and makes only light-bending features such as scratches or steps appear bright.
Related article: Dark-Field Illumination and Edge AI Defect Classification: A New Baseline for Specular-Surface Inspection

Backlight
Lighting from behind the object that captures its outline as a black silhouette. It is mainly used to measure holes and outer dimensions.
Related article: Is a Silhouette Dimension Accurate Just Because the Backlight Is On? — Reducing Edge Blur and Measurement Error on Thick Parts with a Collimated Backlight

Cross-polarization
Placing polarizing filters on the light and the camera at right angles to each other to suppress specular reflection. It also cuts light substantially, so exposure must be recalculated.
Related article: Specular Reflection Erases Defects: How Cross-Polarization Catches Missed Detections on Highly Reflective Surfaces

Strobe
Lighting that turns on for a very short time. That on-time becomes the time over which motion blurs in the image.
Related article: Shorter Exposure Means a Darker Image, So How Do You Freeze Motion? — Suppressing Motion Blur on the Move with Strobe Overdrive

Saturation
The state in which a pixel has hit the brightest value it can hold and cannot distinguish brighter light. It shows up as a blown-out white area.
Related article: Raising the Camera Leaves the Glint of Its Built-in Light in Place

SWIR (short-wave infrared)
Roughly the 900–1,700 nm wavelength band. Ordinary silicon sensors lack sensitivity there, so dedicated sensors are used.
Related article: In SWIR, One Stop Down Erases a 3.4 μm Pixel

Cameras and Sensors

Area-scan camera
A camera whose sensor has pixels laid out in rows and columns and captures a whole frame at once.
Related article: How Much Closer Does a CIS Sit Than a Line Scan Camera?

Line-scan camera
A camera with a single row of pixels that builds an image by stitching lines together as the target moves past.
Related article: How Much Closer Does a CIS Sit Than a Line Scan Camera?

TDI (time delay integration)
A line-scan method that adds up the signal from several lines in step with the target’s motion to lengthen effective exposure. It makes up for light shortfalls on fast lines.
Related article: The Line-Scan Camera You Picked by Line Rate Alone — Why It Still Misses Defects: TDI and the Resolution Design Relationship

Global shutter
Exposing every pixel on the sensor at the same moment. Moving objects keep their shape.
Related article: A Board at 0.5 m/s Leans 15 mm Under a Rolling Shutter

Rolling shutter
Exposing the sensor’s rows one after another from top to bottom. Each row is captured at a different moment, so moving objects can appear skewed.
Related article: A Board at 0.5 m/s Leans 15 mm Under a Rolling Shutter

Bayer pattern
An arrangement that places one red, green or blue filter over each pixel. Because each pixel measures only one color, each color channel has lower resolution than the total pixel count.
Related article: A Board at 0.5 m/s Leans 15 mm Under a Rolling Shutter

Quantum efficiency (QE)
The fraction of incoming photons a sensor converts into electrical signal. It varies with wavelength, so it must be checked whenever the lighting changes.
Related article: Changing the Sensor Means Recalculating the Illumination — What 15% Quantum Efficiency at 940 nm on a BSI Global Shutter Sensor Actually Means

Bandwidth
The number of bits a cable can carry per second. The image data a camera produces per second must stay below it or frames back up.
Related article: Camera Interface Bandwidth Determines Resolution — From GigE Vision to CoaXPress 2.0

GigE Vision and CoaXPress
Video transmission standards for industrial cameras. GigE Vision sends images over Ethernet, and CoaXPress over coaxial cable.
Related article: Camera Interface Bandwidth Determines Resolution — From GigE Vision to CoaXPress 2.0

Smart camera
A camera that combines the sensor, a processor that makes decisions, and I/O in one body.
Related article: How Much Closer Does a CIS Sit Than a Line Scan Camera?

Event camera
A camera in which each pixel sends a signal only when its brightness changes. It can catch motion shorter than a frame interval, but its output differs from ordinary images, so processing must be rebuilt.
Related article: Ultra-High-Speed Micro-Defects Caught by Event Cameras (Neuromorphic Vision) — Seeing What Frames Miss

Decision Algorithms

Rule-based inspection
Judging by numerical conditions people define, such as brightness thresholds, area and length. The condition that triggered becomes the reason for the decision.
Related article: Deep Learning vs. Rule-Based Algorithms: When to Use Which

Deep learning
A multi-layer neural-network approach that learns decision criteria from many example images without people writing rules.
Related article: Deep Learning vs. Rule-Based Algorithms: When to Use Which

Blob analysis
Finding connected groups of pixels in a thresholded image and measuring their area, length and position.
Related article: Deep Learning vs. Rule-Based Algorithms: When to Use Which

Subpixel edge detection
Reading where an edge passes between pixels, more finely than one pixel, by fitting the brightness change of the surrounding pixels.
Related article: Deep Learning vs. Rule-Based Algorithms: When to Use Which

Anomaly detection
Learning from normal images only, then finding areas that deviate from normal. It is common on lines where defect images are hard to collect.
Related article: When the Light Dims by 5 %, How Many Good Pixels Turn Into Defects?

Overdetection and missed detection
Overdetection judges a good part as defective; a missed detection lets a defective part pass as good. Reducing one tends to raise the other, so the line is set by cost.
Related article: Overdetection vs. Missed Detection — Which One Costs More?

Few-shot learning
Building a decision model from only a handful of defect samples. It does not guarantee rare defects never seen before.
Related article: Few-Shot Anomaly Detection: Can 5 Defect Samples Stand Up an Inspection Line

Flat-field correction
Correcting brightness differences across the frame using a reference-card image. Where the correction gain is high, noise grows too.
Related article: Can Correction Gain Erase Uneven Illumination? — The Limits of Flat-Field Correction and Designing Uniformity First

Edge AI and Systems

Edge AI
Running a model directly on a small board next to the machine instead of on a server. It cuts transfer latency but adds limits on compute and heat.
Related article: It Worked Fine on the Server — Why Does Accuracy Drop on the Edge?

INT8 quantization
Converting a model’s computation from 32-bit floating point to 8-bit integers to cut computation and memory. Accuracy on thin-defect classes can drop first.
Related article: What You Must Check Before Applying INT8 Quantization to Edge AI Inference: Why Micro-Defects Quietly Disappear

NPU (neural processing unit)
A processor dedicated to neural-network computation.
Related article: Edge AI Inference Latency: The Camera Interface Decides Before the NPU Does — GigE Vision 3.0 RDMA and Disaggregated Inspection Architecture

Tiled inference
Cutting a large image into pieces the size of the model input and running inference on each. The pieces overlap slightly so defects on boundaries are not missed.
Related article: Where Do Fine Defects Go When You Shrink to 640 × 640? — Overlap, Boundary and Latency Design for Edge AI Tiled Inference

p99 latency
The latency within which 99 % of measurements fall. Even when the average meets the deadline, decisions are occasionally late if this value exceeds it.
Related article: Is It Safe When the Average Latency Fits the Deadline? — Breaking Down the Edge Vision Inspection Latency Budget by p99

Tact time
The time allotted for one part to pass through a process. Capture and decision must finish within it.
Related article: Is It Safe When the Average Latency Fits the Deadline? — Breaking Down the Edge Vision Inspection Latency Budget by p99

For a suggested reading order, see the Start Here guide.

Last updated: October 24, 2026