Specification conversations about inspection cameras tend to collapse into one number: megapixels. It compares neatly across quotations and fits on a datasheet. Yet lines report the same complaint from opposite directions. A high-resolution camera still passes scratches that a lower-resolution system on a neighbouring line catches, while a modest sensor aimed at a small field of view outperforms equipment costing several times as much.
The smallest defect a camera can see is not set by its resolution alone. It is set by the spatial scale of one pixel on the part, which equals the field of view divided by the number of pixels across that axis. Under sampling theory a defect must span at least two pixels to be resolved at all, three to five to be detected reliably, and ten or more when the task is dimensional measurement. Pixel size then decides how much light each sample collects, and whether that theoretical limit survives real optics, lighting and motion.
The sections below convert those two figures into a millimetres-per-pixel scale, then test where sampling, diffraction, contrast and speed narrow the answer.
How Resolution and Pixel Size Describe Different Things
Resolution counts samples; pixel size measures the distance between them. More pixels place samples more finely across the same area, while larger pixels collect more light at each sample. Neither figure tells you what the camera can detect until both are referenced to a field of view and to the defect.
Both specifications sit side by side on every datasheet, which encourages the assumption that they are variations on one theme. One counts measurement points; the other sets the width of the sampling interval. Because requirements run from a presence check to the full set of visual inspection detection functions on a mixed line, no single combination suits every task.
Resolution Is a Pixel Count
Resolution is the number of pixels on the sensor, quoted as a rectangular array. A 2,448 by 2,048 sensor holds just over five million photosites and is sold as 5 megapixels. The total is convenient for comparison and nearly useless for engineering, because a defect running horizontally is limited by the 2,448 pixels on that axis. Five megapixels across a 40 mm field yield 0.016 mm per pixel; across 600 mm they yield 0.245 mm per pixel, fifteen times coarser, with no hardware change.
Pixel Size Is the Sampling Interval
Pixel size is the physical width of one photosite, in micrometres. Multiply it by pixel count and you obtain the sensor dimensions: a 1,024 by 1,024 array with 5.5 micrometre pixels occupies 5.6 mm square, while the same array built from 3.45 micrometre pixels occupies 3.5 mm square. A photosite converts photons in proportion to its area, so the larger pixel gathers about two and a half times the light. Larger pixels buy signal; smaller pixels buy sampling density.
| Property | Camera resolution | Pixel size |
| Unit | pixels | micrometres |
| Describes | how many samples are taken | the interval between samples |
| Increasing it | finer sampling within a fixed sensor area | more light per sample, coarser sampling |
| Effect on smallest visible defect | lowers millimetres per pixel | sets the light budget behind each sample |
| Main cost | bandwidth, frame rate, lens quality | sensitivity, dynamic range, sensor area |
Calculating the Smallest Defect Your Camera Can See
Divide the field of view by the pixel count across that axis to obtain millimetres per pixel, then divide the smallest defect by that figure to see how many pixels it occupies. A defect narrower than two pixels cannot be resolved in principle, and three to five pixels is the practical threshold for dependable detection.
The estimate takes two minutes and eliminates most unsuitable cameras before a quotation is written. Only the field of view, the defect size and the pixel count along one axis are needed.
The Millimetres per Pixel Formula
The calculation is simple enough to run during a quotation review, which is why it is often skipped. Millimetres per pixel equals the field of view on one axis divided by the pixel count on the same axis. If 300 mm must be imaged across 4,096 pixels, each pixel represents 0.073 mm and a 0.3 mm defect occupies roughly four pixels.
Specification work runs the arithmetic backwards. Required pixels equal the field of view divided by the defect size, multiplied by the pixels wanted across the defect. For a 0.3 mm defect in a 300 mm field, targeted at four pixels, the answer is 4,000 pixels on that axis. A 4,096 pixel sensor meets the requirement; a 2,448 pixel sensor does not, whatever its megapixel figure.
How Many Pixels Should Cover a Defect
No universal constant exists, because the answer depends on how much the defect’s appearance varies. A missing component changes a large area abruptly; a hairline scratch on a brushed surface may shift grey level by a few percent. These figures are engineering starting points rather than standards, and each should be confirmed with real samples.
| Inspection task | Pixels across the feature | Reasoning |
| Presence or absence | 2 to 3 | a stable area change is sufficient |
| Surface defect detection | 3 to 5 | contrast varies with finish, angle and lighting |
| Dimensional measurement | 10 or more | tolerance and edge repeatability dominate |
| OCR and code reading | 5 or more per stroke | print quality and distortion consume margin |
| Robot guidance | 5 or more on the fiducial | position repeatability matters more than size |
The margin absorbs variation; it does not improve the optics. A feature occupying exactly one pixel flickers between light and dark as the part shifts, and no classifier stabilises a signal that unstable.
A Worked Example at Three Field of View Sizes
Take one 12 megapixel sensor with a 4,096 pixel array and aim it at three fields of view. The camera never changes; the capability does.
| Field of view | Millimetres per pixel | Two pixel floor | Practical limit at five pixels |
| 60 mm | 0.0146 | 0.029 mm | 0.073 mm |
| 300 mm | 0.073 | 0.146 mm | 0.366 mm |
| 600 mm | 0.146 | 0.293 mm | 0.732 mm |
Doubling the field halves the real resolution, so covering twice the area with the same sensor doubles the defect size that stays detectable. Where a wide field and a fine defect are both mandatory, the answers are more cameras, a narrower field per station, or a line scan arrangement.
Why the Nyquist Limit Sets a Physical Floor
Sampling theory requires at least two samples per cycle to reconstruct a pattern, which in imaging means a defect must be separated from its surroundings by at least one pixel. Features narrower than roughly two pixels stay invisible however the software is tuned.
That floor is geometric rather than electronic, so it applies to every sensor, lens and algorithm combination in the same way.
Sampling, Aliasing and the Two Pixel Rule
Picture two bright features a short distance apart. If the gap allows one pixel to fall entirely inside it and record no signal, the features appear separate. If the gap is narrower than a pixel, that dark sample never occurs, the features merge, and the information disappears before any algorithm sees it. This is the practical form of the Nyquist-Shannon sampling theorem applied to imaging, and it is why small pixels extend capability while interpolation does not.
Near the limit, detail does not fade gracefully. It returns as a false pattern, a moiré band, or an edge that wanders as the part moves. Inspectors sometimes read those artefacts as real defects, which makes an under-sampled system worse than an obviously under-resolved one.
When the Lens Becomes the Bottleneck
The sensor can only record detail the optics deliver. An aperture diffracts light into a spot rather than a point, and the diameter of that spot, the Airy disk, scales with wavelength and f-number. At 550 nm and f/4 it measures about 5.4 micrometres. Fit a 2.4 micrometre pixel behind that lens and each pixel samples a blur wider than itself; the extra pixels record the blur more precisely without revealing more detail. The principle of optical resolution admits no exception: the weakest link sets the limit.
What Is a Good Pixel Size for a Camera
There is no universally correct pixel size, only one appropriate to the light available and the detail required. For general industrial inspection under controlled illumination, 3.45 to 5.5 micrometres is a balanced range; 2.4 micrometres suits high-magnification work on small fields with bright, even lighting; 6 micrometres and above suits low-light or high-speed work where signal quality outranks fine detail.
The choice narrows once three variables are fixed: illumination available, defect contrast expected, and line speed.
| Pixel size | Typical role | Strengths | Constraints |
| Around 2.4 micrometres | high magnification, small field | finest millimetres per pixel | lower full well capacity, demanding light |
| Around 3.45 micrometres | mainstream inspection | balanced detail and signal | lens must match the sampling frequency |
| Around 4.8 micrometres | general purpose, moderate speed | good signal-to-noise, tolerant lighting | larger sensor for the same pixel count |
| 5.5 to 6.5 micrometres | low light, high speed, coarse field | high sensitivity, wide dynamic range | coarser sampling, bulkier optics |
What Is the Smallest Pixel Size
Consumer sensors now reach below one micrometre, and they are impressive for photography but poor models for inspection. The industrial floor sits near 2.2 to 2.4 micrometres, held by three constraints. Full well capacity scales with pixel area, so halving pixel width quarters capacity and compresses dynamic range. Microlenses over tiny photosites accept light efficiently only near normal incidence, so off-axis rays leak into neighbouring pixels as crosstalk and degrade modulation transfer function. Diffraction then stops progress once the Airy disk exceeds a few pixel widths.
The cost appears as noise rather than as a specification failure. A smaller photosite collects fewer photons in the same exposure, so signal-to-noise falls, and exposure must lengthen or light brighten to compensate. On a fast line neither option is available. This is where a resolution decision becomes a throughput decision, and where measured characteristics compared under a framework such as the EMVA 1288 standard matter more than headline pixel counts.
Resolution, Megapixels and Video Standards
No. 1080p is 1,920 by 1,080 pixels, approximately 2.1 megapixels, while a 12 megapixel industrial sensor is typically around 4,096 by 3,000. The larger device carries nearly six times the pixel count and more than twice the horizontal resolution, so the two formats are not comparable for inspection.
The two terms come from different industries and answer different questions.
Is 12MP the Same as 1080p
The confusion arises because both terms are used loosely in procurement, one borrowed from broadcasting and one from sensor marketing. Aspect ratio matters as much as the total: a 12 megapixel sensor in 16:9 carries fewer vertical pixels than one in 4:3, and a defect running vertically is limited by the shorter axis.
| Format | Pixel array | Megapixels | Horizontal pixels |
| VGA | 640 by 480 | 0.3 | 640 |
| 720p | 1,280 by 720 | 0.9 | 1,280 |
| 1080p | 1,920 by 1,080 | 2.1 | 1,920 |
| 5 megapixels | 2,448 by 2,048 | 5.0 | 2,448 |
| 12 megapixels | 4,096 by 3,000 | 12.3 | 4,096 |
What Is the Lowest Camera Resolution
There is no formal minimum, and VGA sensors at 640 by 480 remain in service wherever the task is coarse, such as confirming that a part sits in a fixture. The floor is set by arithmetic, not convention. If a 640 pixel axis covers 200 mm, each pixel represents 0.31 mm and only defects above roughly 1 mm are realistically detectable. Where that suffices, a low-resolution sensor is the correct choice, cheaper and easier to light. Where it does not, no software closes the gap.
Detection and Measurement Need Different Resolutions
Detection asks whether something is present; measurement asks how large it is within a stated tolerance. Detection can succeed with three to five pixels across a feature, while measurement usually needs ten or more, plus calibration, distortion correction and repeatable part presentation.
Buyers often specify a camera for measurement performance and then judge it on detection results, or the reverse. A system that reliably flags a chipped edge may be incapable of reporting chip depth to ten micrometres. For an optical measuring machine, the relevant figure is the uncertainty of the whole chain: lens distortion, perspective error, calibration drift, stage repeatability and edge definition all contribute.
Sub-Pixel Processing Has Limits
Sub-pixel algorithms estimate the position of a well-defined edge to a fraction of a pixel, sometimes a tenth, which is useful for gauging and alignment. It is not a substitute for resolution. An edge that was never resolved cannot be recovered by interpolating between samples that contain no information about it. Sub-pixel processing refines measurement of a visible feature; it does not make an invisible one appear.
What Else Limits the Smallest Visible Defect
Resolution and pixel size define what the sensor can sample, not what the system can see. Contrast, focus, depth of field and motion blur each impose an independent limit, and the tightest of them governs the outcome.
The constraints below decide whether the calculated resolution survives contact with the production line.
Contrast Comes from Lighting, Not Pixels
A camera cannot detect a feature that produces no difference in the light reaching it. A fine scratch on polished metal may vanish under diffuse frontal illumination and appear sharply under low-angle darkfield light, which the scratch scatters toward the lens while the surrounding surface does not. A silhouette against a backlight reveals profile defects that front lighting never shows. Resolution supplies detail; illumination supplies the difference that makes detail meaningful.
Motion Blur and Depth of Field
If a part moves while the shutter is open, the image smears. Blur in pixels equals part velocity multiplied by exposure time, divided by millimetres per pixel. Beyond about one pixel of blur, fine detail erodes and measurement degrades earlier. Short exposures solve blur but reduce collected light, returning the problem to pixel size and illumination. Depth of field then adds its own constraint, because parts that vary in height or rest loosely on a conveyor drift outside the usable focus band, and the finest sampling shrinks that band.
Multi-View Coverage for Complex Parts
A single camera sees one face. Fasteners, connectors and moulded components carry defects on several surfaces at once, and inspecting only the top leaves the rest unexamined regardless of resolution. Multi-view arrangements using several cameras or indexed rotation are the usual answer, which is why six-sided appearance inspection equipment exists: resolution cannot compensate for geometry.
How to Specify Resolution for a Real Inspection Task
Start from the defect and the field of view, not from a megapixel figure. Calculate millimetres per pixel, apply a margin suited to the task, confirm the lens can resolve that scale, verify that the lighting reveals the defect, and only then commit to hardware.
A supplier can size a system quickly when the inputs are complete, and will guess badly when they are not.
Information to Prepare Before Selecting a Camera
- The complete field of view, including positional variation of the part within it.
- The smallest defect that must be rejected, measured on its narrowest dimension.
- The acceptance criterion: detection, classification, or measurement to a stated tolerance.
- The cycle time or line speed, which fixes maximum exposure and required frame rate.
- The working distance and any mechanical constraints on camera placement.
- Representative good, defective and borderline samples that show natural variation.
- Surface finish and colour, which govern lighting strategy and contrast.
With those inputs the arithmetic above produces a defensible specification. Applied to sorting systems, it explains why the resolution figures quoted for an optical sorting machine mean little without its feed rate, field of view per station and the defect classes it must separate. A machine specified at high resolution over a narrow track cannot be compared with one covering a wide belt at half the pixel density.
Most projects then follow a familiar sequence: a calculation that narrows the options, a lens and lighting trial that eliminates some of them, and a feasibility test on real samples that confirms the survivor.
Frequently Asked Questions
Can a 5 megapixel camera detect a 0.05 mm defect?
It can, but only across a narrow field. A sensor with a 2,448 pixel axis, targeted at four pixels across the defect, needs 0.0125 mm per pixel, limiting the field of view to roughly 30 mm. Beyond that width the camera cannot resolve the defect. The camera is capable; the geometry is the constraint.
Does a colour camera lose resolution compared with a monochrome one?
Effective resolution drops. A colour sensor filters each photosite so it records one primary colour, and missing values are interpolated from neighbours. This lowers achievable modulation transfer function and softens fine detail. Monochrome remains the better choice for fine surface work, with colour reserved for tasks where hue carries the discriminating information.
When should a line scan camera replace an area scan camera?
Line scan suits continuous material such as film, sheet, wire or textile, very wide fields where an area sensor would need excessive pixels, and cylindrical parts rotated through the view. Resolution along the scan direction is set by line rate and part speed rather than pixel count, making the motion calculation the critical one.