As an Amazon Associate, Akermin earns from qualifying purchases. Product links on this page are affiliate links.
A super resolution figure is the size of the pixel grid the camera writes out, and in handheld thermal instruments it is usually larger than the microbolometer array that produced it. The detector count is hardware and fixes the angle each measurement covers. The output count is a processing choice and fixes only how many values appear in the file. When a listing names one number and calls it the resolution, the question to ask is which of the two it is, and the giveaway is whether the sentence contains a verb of improvement.
Detector count is hardware, output pixel count is a decision
Inside an uncooled thermal camera is a microbolometer focal plane array: a grid of tiny suspended membranes, each of which warms slightly when infrared energy lands on it and reports that warming as a change in resistance. The number of membranes is set when the wafer is made. It cannot be increased by software, and the physical spacing between their centres, the detector pitch, is likewise fixed in silicon.
Everything downstream of that array is a processing pipeline. The raw values are corrected for non uniformity, offset against a shutter reference, converted to temperature through a radiometric model, mapped to a colour palette and then written into an image of whatever dimensions the designer chose. That final choice has no particular reason to match the array. A display has its own pixel count, a file format has conventions, and a smoother picture is easier to interpret, so resampling to a larger grid is a reasonable thing to do. The trouble is only that the resampled grid is the number that reaches the listing.
The distinction is familiar from other instruments. A meter with a high digit count is not more accurate than its measurement circuitry allows, and a logger that writes a reading every second is not sampling faster than its sensor can respond. In each case a presentation parameter has been mistaken for a measurement parameter. The discussion of what thermal camera resolution actually means works the same separation from the other end.
Interpolation and micro scanning are not the same trick
Two quite different things travel under the super resolution name, and only one of them adds information.
Micro scanning is the real one. A mechanism shifts either the optics or the array by a precise fraction of a detector pitch, typically half, and captures a frame at each of several shifted positions. Because each shifted frame samples the scene at points the others missed, combining them produces a grid with genuinely more independent measurements than the array has elements. It is a well established technique, and it has costs: a moving element in the optical path, several frame times to gather one composite image, and a requirement that nothing in the scene moves during the sequence. Those costs are why it appears in laboratory and fixed installation instruments far more often than in something held in one hand.
Interpolation is the other one. The processor takes the values it already has and computes intermediate values between them by fitting a smooth function through the neighbours. Bicubic resampling and its relatives are standard, fast and free of moving parts, and they produce a pleasant image. What they cannot do is report anything that was not already sampled. Every new pixel is an arithmetic combination of old ones, so the information content of the image is unchanged and only its file dimensions have grown. A claim to boost a thermal image to a larger grid in real time at a video frame rate, with no mention of a scanning mechanism, is describing this second process.
What these four listings publish about their arrays
The clearest way to see the pattern is to line up listings that all quote the same output figure. The Firecore thermal imager states it plainly: a 160 by 120 infrared resolution enhanced by super resolution technology to deliver an effective 320 by 240, together with a frame rate at or below 25 Hz, a field of view of 40 by 30 degrees, a sensitivity below 50 mK, 32 GB of internal storage and a measurement range from minus 4 to 1022 degrees F with an accuracy of plus or minus 2 degrees C or 2 percent over most of it. Both numbers are present, the relationship between them is named, and the field of view is given, which is enough to calculate the angular sampling.
The others state the result without the input. The REVASRI thermal camera says its super resolution technology can improve the image quality to 320 by 240, with a 25 Hz refresh, a 40 mK sensitivity, nine colour modes and fifteen times electronic zoom. The Raythink EX10 Pro says it boosts thermal resolution to 320 by 240 in real time on a 2.8 inch display, with a 40 mK sensitivity, four imaging modes from a 2 megapixel visible camera and six selectable measurement spots. The DCTOPT thermal camera is the most tangled of the four: its title pairs a 320 by 240 display with a 240 by 180 super infrared resolution, while a bullet describes a 2.8 inch screen paired with 240 by 240 super resolution technology, alongside a sensitivity below 50 mK and a 25 Hz frame rate.
| Listing | Output figure as stated | Native array | Sensitivity | Field of view |
|---|---|---|---|---|
| Firecore thermal imager | Effective 320 by 240 | 160 by 120, stated | Below 50 mK | 40 by 30 degrees |
| REVASRI thermal camera | Improved to 320 by 240 | Not stated in the listing bullets | 40 mK | Not stated in the listing bullets |
| Raythink EX10 Pro | Boosted to 320 by 240 | Not stated in the listing bullets | 40 mK | Not stated in the listing bullets |
| DCTOPT thermal camera | 320 by 240 screen, 240 by 180 and 240 by 240 both named as super resolution | Not stated in the listing bullets | Below 50 mK | Not stated in the listing bullets |
Three of the four have made the arithmetic in the next section impossible to perform from the published information. That is the practical cost of the omission, and it is why a side by side method such as the one in comparing resolution between thermal cameras has to start by identifying which number is which.
The angle one detector sees does not move when pixels are added
Take the one listing that gives both numbers and work it through. A 160 element row spread across a 40 degree horizontal field gives 0.25 degrees per detector, which is 4.36 milliradians. At a target distance of one metre that detector covers a strip 4.36 mm wide. Reliable spot temperature measurement conventionally needs the target to fill several detectors across, commonly taken as three, so the smallest object you can measure confidently at one metre is around 13 mm across.
Now repeat the calculation believing the output figure. A 320 element row across the same 40 degrees gives 0.125 degrees, 2.18 milliradians, 2.18 mm at one metre and a measurable object of about 6.5 mm. That answer is wrong by a factor of two in every dimension, and it is wrong in the dangerous direction: it tells you that a small terminal, a fuse clip or a surface mounted component is measurable when it is not. The scene is unchanged, the optics are unchanged, the detector pitch is unchanged, and only the file dimensions moved.
This is also why sensitivity and resolution answer different questions. A 40 mK sensitivity describes how small a temperature difference one detector can distinguish from its own noise when the target fills its cone. It says nothing about whether the target fills the cone. The two figures have to be read together, and the same pairing governs what a usable thermal image for building inspection needs, where large uniform surfaces forgive a small array and small components do not.
Reading a super resolution claim back to its detector
Three habits recover most of the missing information. First, read the verbs: boosts, improves, enhances, delivers an effective and upgrades all describe a transformation and imply an input the sentence has not named. A flat statement with no such verb is usually the array. Second, look for two numbers in one sentence, because a maker who names both has told you the relationship and the smaller number is the one that matters. Third, when only one number appears, treat it as the output until the manual or the specification page says otherwise, and remember that the common upscaling step doubles each dimension, so the array behind a stated 320 by 240 is frequently 160 by 120.
Then check what the camera will give you afterwards. A file that stores the measured values rather than a coloured picture lets you re examine a scene later and see how many independent samples really sit on the object of interest, which is the argument for cameras that save radiometric image files. If the work genuinely needs small targets resolved at a distance, the array has to grow rather than the output grid, and the options for that are set out in the survey of thermal cameras with larger detector arrays. A single honest sentence naming both numbers is worth more than any amount of processing applied afterwards.
Frequently Asked Questions
Is super resolution always just interpolation?
No, but in handheld thermal cameras at this level it usually is. Genuine super resolution by micro scanning shifts the optics or the detector by a fraction of a detector pitch, captures several frames at the shifted positions and combines them, which really does add measurements rather than inventing them. It needs a moving element, a stable scene and time, so it works poorly on a handheld instrument pointed at a changing target. When a listing claims a boosted figure in real time at a video frame rate, interpolation is the likely mechanism.
Why would a maker upscale at all if it adds no detail?
Because a smoothly interpolated image is genuinely more pleasant to look at and easier to interpret than a blocky one, and because the output grid has to match a display and a file format anyway. Interpolation between samples is a legitimate display technique. The problem is not the processing, it is quoting the processed grid as the resolution without naming the array, which invites the buyer to infer a measurement capability that is not there.
How do I tell an output figure from an array figure?
Look for verbs. Phrases such as boosts to, improves to, enhanced to, effective and super resolution all describe a transformation, which implies a smaller input somewhere. A plain array figure is stated flatly with no verb of improvement attached. The strongest signal is a listing that names two numbers in the same sentence, because the smaller one is the detector and the larger one is the output.
Does the thermal sensitivity figure tell me the array size?
Not directly, although the two are related through detector design. Sensitivity is quoted as a noise equivalent temperature difference in millikelvin and describes the smallest temperature change one detector can distinguish from its own noise. It is a property of each element, not of how many of them there are. A camera can pair a very good sensitivity figure with a small array, and that combination is quite common, because fewer and larger detectors collect more energy each.
Does an upscaled image affect temperature measurement accuracy?
It affects which readings you can trust more than it affects the stated accuracy figure. The accuracy quoted in a listing applies to a target that fills the measurement spot comfortably. Upscaling produces apparently fine detail on small objects, and a spot measurement placed on one of those interpolated pixels is reading a value computed from neighbouring detectors rather than a measurement of that point. The number on screen looks precise and is not a measurement of what the crosshair is sitting on.
Is a camera with a stated array always better than one without?
Not better in every respect, but better documented, and for a measuring instrument that counts for a lot. A listing that publishes both the array and the processed figure has told you what it is doing and allowed you to compute the angular sampling for yourself. A listing that publishes only the processed figure has made that calculation impossible from the outside, and it is the calculation that determines whether the camera can resolve the thing you bought it to look at.