As an Amazon Associate, Akermin earns from qualifying purchases. Product links on this page are affiliate links.
They disagree because they are referenced to different things and nothing obliges them to match. Resolution is the smallest change an instrument can display or discriminate. Accuracy is how close the displayed value is to the true one. An instrument that resolves 0.1 units while guaranteeing plus or minus 2.5 units is out by a factor of twenty five, and it is not contradicting itself: the fine number says what it can see change, the coarse number says where the whole scale might be sitting. On real thermal imaging listings the gap runs wider than twenty five, reaching forty, fifty and beyond.
Resolution, sensitivity and accuracy answer three different questions
Four quantities get mixed together on specification sheets, and separating them removes most of the confusion. Resolution is a display property: the size of the least significant step the instrument can show, set by the conversion and counting arrangement behind the screen. Sensitivity, which thermal imaging expresses as noise equivalent temperature difference in millikelvin, is a discrimination property: the smallest difference that rises above the instrument’s own noise, so it governs whether two things in one observation can be told apart. Accuracy is an anchoring property: how far the indicated value sits from the true value, degraded by assumptions about the surface, by radiation arriving from elsewhere, by distance and by the instrument’s own thermal state. Repeatability is a consistency property: the spread across repeated observations of something unchanged.
Only accuracy requires a traceable reference. The other three are internal to the instrument, which is why they can be far finer. A camera with excellent noise performance and a modest accuracy band will show you exactly which terminal in a row is running hot and will not tell you reliably how hot it is. That is a tremendously useful instrument for finding faults and a poor one for writing down a temperature, and the specification sheet says so clearly once you know which number is which. The general framework is set out in how to compare sensor accuracy, resolution and repeatability.
Where the factor of twenty five comes from
The ratio is a plain division: the accuracy band divided by the resolution or discrimination step. An instrument that steps in tenths and guarantees two and a half units gives twenty five. Published figures on real instruments produce larger numbers than that.
| Smallest step the instrument can show or discriminate | Published absolute accuracy band | Ratio between them |
|---|---|---|
| 0.1 unit of display resolution | Plus or minus 2.5 units | 25 times |
| Under 50 mK of thermal sensitivity | Plus or minus 2 degrees Celsius | About 40 times |
| 40 mK of thermal sensitivity | Plus or minus 2 degrees Celsius | 50 times |
| 0.035 degrees Celsius of discrimination | Plus or minus 2 degrees Celsius | About 57 times |
| 0.035 degrees Celsius of discrimination | 2 percent of a 300 degree reading, so 6 degrees | About 170 times |
| 10 microamps of display resolution | Not stated in the listing bullets | Cannot be computed |
The fifth row is where the arithmetic turns nasty. Accuracy stated as a fixed figure or a percentage, whichever is greater, means the band stops being constant once the percentage overtakes the floor. With a two degree floor and a two percent slope, that crossover sits at a hundred degrees, and by three hundred degrees the band has tripled while the discrimination has not changed at all. The ratio is therefore not a single number for an instrument, it is a curve that steepens as the reading rises.
Thermal sensitivity against absolute accuracy on real listings
The published pairs make the pattern concrete. The Thermal Master P3 states that it detects temperature differences as small as 0.035 degrees Celsius, upscales a native 256 by 192 detector to 512 by 384, refreshes at 25 Hz, focuses manually down to 8 mm for circuit board work, offers adjustable emissivity, and measures from minus 20 to 600 degrees Celsius with an accuracy of plus or minus 2 degrees Celsius. Divide one by the other and the two figures differ by more than fifty times.
The REVASRI thermal imaging camera states a 40 mK thermal sensitivity, a 25 Hz refresh, a 320 by 240 super resolution image, nine colour palettes, a fifteen times digital zoom and a reading range from minus 20 to 450 degrees Celsius with a maximum error of 2 degrees Celsius or 2 percent. The division gives fifty at the floor. Neither listing is being evasive. Both are publishing exactly the two quantities a thermographer needs, and both invite the mistake of reading the smaller one as a promise about the larger.
Emissivity is what keeps the absolute figure honest. The camera measures radiation and converts it to a temperature using an assumed emissivity, and a surface that emits less than assumed looks cooler while the deficit is filled by radiation reflected from the surroundings. An error of a few hundredths in that assumption can shift the indicated temperature by an amount comparable to the entire accuracy band, and no amount of noise performance helps, because the noise was never the limiting factor. This is also why adjustable emissivity on a listing is a more meaningful feature than another thousand interpolated pixels, and why thermal figures are reported the way they are in thermal camera resolution explained.
A listing that disagrees with itself about resolution
Sometimes the inconsistency is inside one page. The DCTOPT thermal camera carries three different resolution figures in the same listing: a 320 by 240 figure and a 240 by 180 super resolution figure in the product name, and a 240 by 240 super resolution claim in the feature bullets. The native detector array is not stated separately anywhere. The same listing publishes a thermal sensitivity under 50 mK, a 25 Hz frame rate, a 2.8 inch screen, a 4 GB removable card, a range from minus 20 to 600 degrees Celsius and an accuracy of 2 degrees Celsius or 2 percent.
Three incompatible numbers for the same property is a useful warning about how listing figures are produced. It also exposes the interpolation question. Super resolution is an upscaling step that produces more displayed pixels than the detector has, and it adds no thermal information, so the figure that governs what the camera can genuinely separate is the native array. A listing that publishes both numbers, as the Thermal Master page does when it names 256 by 192 native and 512 by 384 enhanced, is telling you more than one that publishes only the larger figure, even though the larger figure looks better. The distinction between a span, a step and a true separable detail is the same one made in measurement range against resolution and in resolution in a digital meter.
Reading a specification sheet without being misled by the smallest number
Go through the numbers and label each one first: display step, discrimination, absolute band or repeatability. Then find the native detector or sensing element figure rather than the processed one. Then read the accuracy statement carefully enough to know whether it is a fixed band, a percentage, or the greater of the two, and compute it at the reading you care about rather than at the floor. Finally, ask what the output is for. A difference, a trend or a location needs discrimination. A number that goes into a report needs the accuracy band, and the band is the limit of what you may claim no matter how many digits the screen shows.
When the accuracy is simply absent, the fine figure is unanchored. The EITAI leakage current clamp meter publishes a 10 microamp resolution across a span from 0.00 mA to 300 A, a 40 mm jaw, a 200 gram body, peak and data hold and storage for ninety nine groups of readings, and no accuracy figure at all. Notice too that 300 A against a 10 microamp step is a ratio of thirty million to one, which no single range delivers, so the resolution figure necessarily belongs to the lowest range only. A fine step with no error band and no range breakdown beside it describes the display and not the measurement, which is the same conclusion reached in carbon dioxide monitor accuracy.
Frequently Asked Questions
Is a fine resolution figure useless then?
Not at all, it is just answering a different question. Fine resolution and low noise let you see that one spot is warmer than its neighbour, that a reading is rising, or that two conductors in a bundle are not sharing current equally. Every one of those is a differential judgement within a single observation, and the absolute accuracy never enters it.
What is super resolution on a thermal camera?
It is an interpolation that produces more displayed pixels than the detector has, usually by combining slightly shifted frames or by upscaling. The extra pixels make edges easier to read on screen and add no new thermal information. The figure that limits what the camera can genuinely separate is the native detector array, which is why a listing that publishes both numbers is being more useful than one publishing only the larger.
Why does emissivity matter so much to absolute accuracy?
Because the camera measures radiation and infers temperature, and emissivity is the factor relating the two. A surface that emits less than assumed appears cooler than it is, and the shortfall is made up by reflected radiation from the surroundings, which the camera cannot distinguish. Getting emissivity wrong by a small amount can move the indicated temperature by as much as the entire accuracy specification.
What does an accuracy stated as a fixed figure or a percentage, whichever is greater, actually mean?
It means the band is the larger of the two at any reading, so the fixed figure governs at low temperatures and the percentage takes over once the reading is high enough that the percentage exceeds it. With a two degree floor and a two percent slope, the crossover sits at one hundred degrees, and above that the band widens steadily while the discrimination stays the same.
Does repeatability tell me anything the other two do not?
Yes, and it is the quantity most often missing. Repeatability is the spread you get from measuring the same unchanged thing several times, so it describes whether a difference between two of your own readings is real. An instrument can be repeatable and inaccurate, which is the ideal combination for tracking change and the wrong one for reporting a value.
How should I choose between a fine resolution instrument and an accurate one?
By what the number is for. If the output is a comparison, a trend or a location, buy discrimination and low noise. If the output is a figure that goes in a report, on a certificate or into a compliance decision, the accuracy band is the limit of what you can claim and resolution beyond it is decoration.