FLIR SC660 vs Therm-App Pro: a thoroughly unscientific comparison

[Ultrapurple]

I thought I would have a quick play with my new-to-me FLIR SC660 top-end (but discontinued) 640x480 camera vs my much cheaper Therm-App Pro, also 640x480.

First, I want to make it quite clear that this is a thoroughly unscientific test. The images were all made hand-held, meaning that although subjects are the same they aren't taken from the same angle or even distance. The FLIR has a 40mm lens of as-yet-unknown aperture; all the Therm-App Pro images were taken using the standard 19mm f/1.1 lens.

The big, big difference between the cameras is that the FLIR is fully radiometric and the Therm-App Pro is not. (In fairness, it doesn’t pretend to be). The FLIR is designed for making scientific measurements, whereas the Therm-App Pro is really only about making pretty pictures. So although the resolution and NETD (sensitivity) figures for the two devices are similar, that’s really where the similarity ends.

I should say that I'm very familiar with using the Therm-App series of cameras but have only had the FLIR for a couple of days. I haven't used any 'tricks' to make either camera look better or worse; as far as I can tell it's a reasonably fair comparison, subject to the notes herein. The Therm-App Pro was 'driven' by the non-stock app, ThermViewer, which generally produces slightly better images than the stock app and includes a calibration routine rather like a manual version of the NUC shutter in the FLIR. I did use this option, which results in marginally better images than the stock app.

ThermViewer uses real-time superresolution to increase the native 640x480 images to 1280x960. Therefore I have re-sized all of the Therm-App images by 50% to 640x480. I will post the large versions separately.

This first image isn't actually a comparison: it's what I saw when I put the Therm-App up close against the FLIR lens. The speckling is, I presume, dust within the optical system. There are splodges on many of the FLIR images that could well be caused by these bit of dust; when I find out how to remove the lens I'll give it a whoosh with the air duster.

[Ultrapurple]

On to the comparisons now. It should be obvious by the watermark which images are from the FLIR camera and which from the Therm-App Pro.

These images show three TV-related boxes – I think there’s a couple of DVD players and a terrestrial digital TV receiver. There’s not much to choose between the images – the Therm-App Pro has given a more ‘contrasty’ result but that’s neither here nor there. Perhaps the Therm-App Pro has a little more contrast on the front panel of the top box, but it’s a close-run thing.

[Ultrapurple]

A DECT phone. Apart from showing that the ThermViewer palette is jazzier than the FLIR and the Therm-App Pro is a little noisier, there’s not much to choose between them.

[Ultrapurple]

A cooker hood. Subjectively, I prefer the ‘look’ of the Therm-App, simply because it has blacker blacks and a larger depth of focus. However, in visible light photography I like using shallow DOF (for effect) so in real-life use I would probably prefer to use the FLIR for things like portraiture.

Sometimes I really wonder what planet I’m on. Comparing thermal cameras on the basis of their potential for portraiture?

[Ultrapurple]

Gas flames. This is where the limitations of the Therm-App Pro really show themselves. The FLIR images are excellent, whereas the hotter bits of the Therm-App Pro image have no definition whatsoever. In the red-and-blue image you can clearly see how the Therm-App Pro has peaked out, leaving only a non-uniform pattern in place of any detail. (The Therm-App Pro is not specified for very hot subjects – in fact, the technical specs don’t actually mention its maximum and minimum sense temperatures. The regular 384x288 Therm-App specs say that version is calibrated for a temperature range of 5-90°C, whereas the SC660 is rated for accurate measurements from -40 to +2000°C). (I understand that the standard SC660 tops out at 1500°C; mine has the optional 2000°C extension).

The FLIR SC660 is a clear winner here.

[Ultrapurple]

This is a test of the low thermal contrast performance. The subject is a ‘stuffie’ (stuffed toy) – in the form of a sloth, who we call Sidney. He has an interesting back-story, but that will have to wait for another time.

Neither camera gives a particularly high contrast image in ‘normal’ mode because the temperature span is tiny. The first two image show what you get ‘out of the box’and I don’t think there’s much to choose between them at first glance.

The second two images are processed versions. I used Paint Shop Pro’s Curves tool to increase the contrast, then the Clarify tool to increase differential contrast a bit. The  difference is quite stark. The Therm-App Pro image responds fairly well to the contrast enhancement but the image is very noisy. The FLIR, on the other hand, is MUCH less noisy when treated in exactly the same way. (It looks as though the focus was a bit off on the FLIR images, for which I apologise).

I think the circular splodges on the FLIR image may be related to the dust specks I mentioned in the first post, but I'm not sure. I have similar marks (at low thermal contrast) on the Therm-App Pro, which I believe may be due to internal reflections present when the camera was calibrated.

The FLIR SC660 wins.

[Ultrapurple]

Summary

Whilst this has been a thoroughly unscientific test, it does prove that the FLIR SC660 is a pretty darn good thermal camera and, thanks to its radiometric images, able to provide a vast amount of data on what it’s seeing. That alone would make it the winner, but in difficult imaging conditions it also clearly out-performs the much cheaper (but newer) Therm-App Pro. Remember, the SC660 was set to full-automatic (including autofocus) for all of these tests. I'm sure I will be able to get better results once I know more about driving it.

On the other hand, the Therm-App Pro has acquitted itself remarkably well. It’s under a tenth of the list price of the FLIR and yet in general imaging use it really keeps up fairly well with the more expensive competition. I was genuinely surprised how well it performed in my more or less like-for-like tests. But of course it’s not radiometric, and is really a domestic-grade device, whereas the FLIR SC660 is a scientific instrument.

A spec sheet for the SC660 can be found here, among other places

and the Therm-App Pro’s product page is here.

I have been unable to find any support for the SC600-series cameras on the FLIR site, which surprises me. Has anyone got a magic URL I could try? (I did eventually find FLIR's original brochure (PDF file) for the SC600 series).

The image below shows the Therm-App Pro as see by the FLIR SC660.

[Fraser]

Here ...

https://flir.custhelp.com/app/account/fl_downloads

Look in FLIR Legacy section for SC6xx handheld cameras.

 

Manual.....

https://support.flir.com/DocDownload/Assets/dl/1558550$a557.pdf

Getting Started Guide....

https://support.flir.com/DocDownload/Assets/dl/t559019$007.pdf

[Vipitis]

Thanks for this initial comparison.

I believe you can find a lot of interesting tricks in the E4 thread. There will be a collection somewhere of all the php scripts to use with ExifTool to get raw images, meaning no judgement elements or water mark.

[eKretz]

Nice, good read. Another thing to consider, which I've noticed on my FLIR Exx and I can definitely also see in your sample images - the cameras with focus adjustment have considerably lower DOF, so their images won't be as sharp throughout the depth of the image, especially with closer subjects. The tradeoff for the increased DOF on the lower end cameras is the higher noise. The contrasty thing is more of a personal preference, but the main thing I don't like about that is the loss of detail. You can lose a lot at either end of the scale that way. I'd rather have all the info and choose what to get rid of rather than have the camera make that decision. I definitely see some more low scale information in some of the FLIR images you've posted than in the TA images.

[Hyper_Spectral]

When you have a moment, I'd be interested in seeing the "super resolution" images. Specifically, I'd like to see how accurate the algorithm that "doubles" the pixels can be.

[Ultrapurple]

When you have a moment, I'd be interested in seeing the "super resolution" images. Specifically, I'd like to see how accurate the algorithm that "doubles" the pixels can be.

Bear in mind I am making no specific claims here, merely that it's what it says on the tin. Subjectively, the superresolution results I have seen appear significantly better than material from the same source without superresolution, but I have never attempted any measurements.

Here's an example video with a 640x480 original post-processed to 1280x960 using Video Enhance 2.

Here is a copy of the as-shot 640x480 original.

Here's an example video superresolution enhanced 'in-camera' by ThermViewer. Please read the notes that accompany each video.

[eKretz]

Wow, that is quite an improvement over the original footage.

[Hyper_Spectral]

Thank you, I didn't realize those videos were yours!

Superficially the improvement is obvious. When it comes to actual accuracy.... indeed, we do need actual measurements

One thing to pay attention to that comes to mind is the recent article posted by one of the large leaders in image based machine learning (I think it was google, referencing satellite imagery). The article explained that while the image produced by the code, to the human eye, appeared to be beautiful and accurately produced, closer examination showed the algorithm just placed chunks of any images pixels where it was trained to, by the human. It would even take pieces of imagery from other locations entirely and obfuscate and merge them into the "produced" image.

Now, I don't believe any machine learning is involved here, but it's worth noting that it's important to distinguish between the use of "real" pixels and "manufactured" pixels.

[eKretz]

That is certainly true for the scientific instrument aspect, but the Therm-App doesn't do radiometric measurement, and in this use case it's more of an artistic instrument, for which the resolution improvement is all benefit and no harm as far as I can see.

[Ultrapurple]

Although the regular Therm-App app Pro aren't sold as radiometric cameras, it is possible (easy) to extract the value of each pixel, either as a CSV file or other similar methods. From memory  I believe the data comprises the scene temperatures in kelvin, to one or two decimals. (I can probably look this up or provide some sample files for someone clever to look at).

The cameras are not well calibrated and there is no temperature stabilisation or NUC flag, so they drift. But you CAN get at the data.

There is a 'thermography' version of the 384x288 Therm-App but I know little about it. I did see a review on YouTube where a 'professional' thermographer set out to trash it (and largely succeeded) but anyone familiar with the basic principles of thermal camera calibration wouldn't get more than a couple of minutes into it before the smell of rats became too overpowering. Example: black body on test bench. Posh thermal camera with (say) 40mm / 24° lens at just the right distance for good, screen-filling results. Cheaper camera with 6.8mm / 50° lens 'placed at the same distance for a fair test' and then electronically zoomed to fill the screen. Cue bitter complaints about the fuzzy, low resolution image (even though the cheap camera has many more pixels than the posh one). I suspect you get the picture; it was downhill all the way from there. The main thing it told me was that the 'professional' thermographer was running scared that his highly profitable buildings inspection business was under threat from cameras that cost less than one of his home surveys.

As far as I know, wedding photographers haven't been put out of business because everyone has a cell phone camera and Uncle Pete has just bought his first SLR - there may have been a bit of that to start with but the population has learned there's more to good wedding photos than point-and-shoot. The same is true of any other profession, including thermography: there's more to it than just the gear. Give Eric Clapton a cheap guitar and he'll still sound like Eric Clapton. Does anyone care what word processor Stephen King used? I could sit at his desk, with his computer, and I doubt very much that I would come up.with a bestseller. (The two books I've had published are on technical matters, which saves all that messing around with plots and characters!)