You are here: Home - Site Information: FAQ, Glossary, Privacy, Terms | EEK - Visual Inspection Glossary

Visual Inspection Glossary

Source:Site Information: FAQ, Glossary, Privacy, Terms | EEK / Date:2026-09-22

This page explains, item by item, the technical terms that appear in this site's technical articles, so you can align terminology during selection discussions.

Note that the following are industry-general meanings; for your specific workpiece, the same word may correspond to completely different acceptance criteria — for example, "accuracy" in presence/absence inspection means whether the smallest feature can be resolved, while in dimensional measurement it means measurement error. If you are not sure which definition to use, send us photos of the workpiece and we will look at them together.

Inspection Task

presence/absence inspection

Judging whether a feature, part or marking exists, and outputting OK / NG. This is the largest category of tasks in industrial vision; typical objects include screws, clips, labels, instruction manuals, connectors and glue beads.

missing part

A part that should be fitted has not been installed. In inspection terms this is "should be present but is absent", and it is usually judged by comparing regional grayscale or contour differences against the standard assembly state.

wrong part

A part is installed, but the model, specification or orientation is wrong. This often requires reading characters, comparing colors or comparing contours to distinguish.

Assembly Seating Inspection

Judging whether a part is seated in its designed position, for example whether a clip is fully engaged or a connector is pushed all the way in. The position offset or gap width of a local feature is commonly used for the judgement.

count inspection

Counting the number of targets in the field of view, such as the number of screws in a bag of fittings. When targets overlap, segmentation is usually needed first; estimating by area alone tends to fail in overlapping scenarios.

Front/Back Inspection

Judges whether the workpiece orientation is correct, for example the front and back sides of a gasket or the direction of a terminal. Asymmetric features or character orientation are often used to distinguish them.

Appearance Defect Inspection

Detecting surface anomalies such as scratches, dirt, burrs, pits, color difference and bubbles. Detection difficulty depends heavily on the contrast between defect and background, so the illumination plan is often more critical than the algorithm.

dimensional measurement

Measures geometric quantities such as length, diameter, angle and hole spacing, and converts them into physical units. Accuracy depends on pixel scale and calibration quality, which is not the same thing as "being able to see it clearly".

Character Recognition (OCR)

Reads text, batch numbers and dates on workpieces or packaging. When the print is clear, traditional template matching can be used; when fonts vary and backgrounds are complex, deep learning is generally required.

Barcode / QR Code Recognition

Reads 1D barcode or QR code content. Beyond the read rate, attention must also be paid to stability with damaged codes, low-contrast codes and motion blur.

Optics and Imaging

Area-Scan Camera

A camera that produces a full 2D image in a single exposure, suitable for stationary or indexing stations. Selection depends on resolution, frame rate, sensor size, and interface (GigE / USB3 / Camera Link).

Line-Scan Camera

A camera that scans line by line and builds an image as the object moves, suited to continuously moving roll material, strip material and cylindrical surfaces. It must be strictly synchronized with the production line speed and places higher demands on light source brightness.

Field of View (FOV)

The actual area the camera can cover on the workpiece plane. FOV is directly related to accuracy: at the same resolution, the larger the FOV, the larger the actual size each pixel represents, and the larger the smallest resolvable defect.

Working Distance (WD)

The distance from the front of the lens to the workpiece surface. It must be calculated together with the FOV when selecting a lens; where site space is insufficient, a shorter focal length lens or an angled optical path is often needed.

Pixel Equivalent

The real-world size on the workpiece that one pixel corresponds to (for example, 0.05 mm/pixel). It is obtained by dividing the FOV by the camera's pixel count in one direction. It sets the theoretical resolution limit, but it is not the actual inspection accuracy.

Depth of Field (DOF)

The range of height variation a workpiece may have along the optical axis while staying sharp. High magnification and a wide aperture make the depth of field shallower, so parts with height differences easily go out of focus locally.

Exposure Time

The time over which the sensor accumulates light. A longer exposure makes the image brighter, but a moving workpiece smears; production lines with a fast cycle time must first shorten the exposure and then compensate brightness with the light source.

Illumination (Light Source Design)

The angle of incidence, color and uniformity of light are used to "bring out" defects. Change the illumination on the same workpiece and defect contrast can go from invisible to immediately obvious.

Ring Light

Illumination directed obliquely from around the lens; the most versatile option, suitable for surface defect and presence/absence inspection of most flat workpieces.

Coaxial Light

Light passes through a beam splitter to illuminate the workpiece perpendicularly; suitable for character and scratch inspection on highly reflective, mirror-like surfaces, and suppresses reflection interference.

Backlight

The light source is placed behind the workpiece so its contour appears as a silhouette. Suitable for dimensional measurement, hole position inspection, and contour integrity judgement, with the highest edge contrast.

Bar Light

A linear light source, often used in pairs to graze the surface from both sides, which highlights directional features such as scratches, press marks and brushed texture.

Dome Light (Dome Illumination)

A dome light source with diffuse reflection on its inner wall emits extremely uniform light; suitable for curved, reflective and complex-shaped workpieces, and eliminates local highlights.

Photometric Stereo

Light from several directions is used to image separately, and the bright-dark variation is then used to infer the surface normal, which can reveal extremely shallow pits and indentations; it is often used for low-contrast surface defects.

Algorithm and Judgement

Threshold Segmentation

Separating the image into target and background by gray level is the most basic positioning and extraction method. It is fast and reliable when illumination is stable; when illumination fluctuates widely it must be combined with normalization or replaced by learning-based methods.

Blob Analysis

Treats a connected pixel region as a single whole and measures features such as area, centroid, length and width, and roundness; commonly used for presence/absence inspection, counting and positioning.

template matching

Use a standard image as a template to search the image under inspection for the most similar position, for positioning and shape comparison. It is sensitive to rotation, scaling and occlusion, so shape matching or multiple templates are required when necessary.

Edge Detection

Extracting positions where the gray level changes abruptly in the image, used to measure edge-to-edge distances and to judge contour integrity. Sub-pixel edges can improve positioning accuracy to a fraction of a pixel.

Deep Learning (CNN)

Using a convolutional neural network to learn defect features directly from samples, suited to defects whose morphology varies and is hard to describe with rules. The cost is the need for a sufficient number of representative samples.

Anomaly Detection

Models only on normal samples and judges any deviation from normal as an anomaly. It suits scenarios where defective samples are extremely hard to collect, but may raise false calls on "normal variation never seen before".

Data Annotation

Defect positions and categories are marked on the training samples. Annotation consistency directly determines the model's ceiling, so the judging criteria for the same defect type must be unified first.

Over-Detection / False Detection (False Positives)

Judging a good part as a defective part. A high over-rejection rate causes frequent line stops, which in actual production often affects usability more than escapes.

Escape (False Negative)

Judging a defective part as conforming. This is the main source of quality risk, and acceptance criteria usually set stricter requirements on the escape rate.

Recall and Accuracy

Recall is the proportion of true defects that are detected, and precision is the proportion of items judged defective that are truly defective. The two must be balanced against the line stoppage frequency acceptable on site, rather than pursuing either one alone.

System and Production Line

Cycle Time (CT)

The total time required to complete the inspection of one workpiece, including loading, image capture, computation, judgement and sorting actions. Cycle time is the first constraint of a solution; usually the cycle time is set first and the camera and light source configuration derived from it.

OK / NG

The judgement result of good or defective. The system must output this signal and drive sorting or an alarm.

PLC

Programmable logic controller, the core of production line control. The vision system generally exchanges results and trigger signals with it through I/O or an industrial bus.

I/O Signals

The simplest integration method: the vision system outputs discrete signals such as OK / NG / complete, and the PLC provides the trigger signal. Wiring is simple and latency is low, but the amount of information that can be transferred is limited.

Profinet / EtherNet-IP / Modbus

Common industrial Ethernet and fieldbus protocols. Through these protocols, judgement results, measured values and even defect image indexes can be transmitted to the PLC or host computer.

MES

Manufacturing Execution System. The vision system can upload inspection results, timestamps and workpiece serial numbers to achieve piece-level traceability.

calibration

The process of establishing the correspondence between image pixels and actual physical dimensions (or robot coordinates). Inaccurate calibration causes systematic bias in measured values.

3D Vision / Point Cloud

Three-dimensional workpiece data is obtained by structured light, laser triangulation or stereo vision, and is used for items that are difficult to judge from two-dimensional images, such as height, flatness, volume and assembly gap.

Need More Specific Explanations?

Call 13913763268, or leave a message via the Contact Us leave a message on the page and we can convert the relevant terms into executable, specific metrics for your workpiece.

Previous: Frequently Asked Questions