Contrary to popular belief, FLIR One IR image does NOT add noise

[Ben321]

If anything, it filters out noise. If you use the sample FLIR One app that came with the SDK you will be able to access the true raw 14bits-per-pixel image. This is unprocessed, directly from the sensor. If you point it at a surface with nearly uniform temperature, it will adjust the min and max levels so that it will try to bring out a lot of detail in something that doesn't have much. You will notice a SIGNIFICANT amount of noise in the image here. However if you look at the processed IR image, you will notice almost all the noise is gone. Instead you may see some things that look like compression artifacts, but I'm guessing that has more to do with the noise-reduction algorithm it uses. I'm guessing that it does a DCT transform like with JPEG, on each 8x8 block of pixels, filters out the higher frequencies (these often will contain more noise than lower frequencies) and then performs the inverseDCT to get back the now-filtered block of pixels. Of course this introduces some artifacts that look like a ripple, as it does any time you use frequency-domain processing (such as FFT or DCT) to filter out certain spacial frequencies in an image.


This whole idea that FLIR intentionally crippled their own product, is pretty silly actually. It's just a reactionary statement that somebody came up with because they didn't like the fact that there was more noise in their image than they would have liked, and sadly (since I've seen more than one post here in which it has been mentioned) it seems like everybody here has started to believe this rumor. Of course, given that this only costs $250, I wouldn't expect its imaging sensor to have gone through nearly as much QC (quality control) at the factory, as would a sensor intended to be installed on a $2000 device. Of course this is how they get their cost down (which is good, as most people aren't going to shell out over $1000 for a device that generates such a small image), and as a result there are of course small imperfections in the vanadium oxide LWIR sensor chip that introduce some noise. Of course this noise then has to be filtered out in either the on-board firmware or in an app (or the "black box" that is the FLIR One SDK itself that you have to reference in your app), or in post processing software (such as FLIR Tools).

[tomas123]

A new thread for an old topic...

This whole idea that FLIR intentionally crippled their own product, is pretty silly actually. It's just a reactionary statement that somebody came up with because they didn't like the fact that there was more noise in their image than they would have liked, and sadly (since I've seen more than one post here in which it has been mentioned) it seems like everybody here has started to believe this rumor.

Do you know, that all Flir Cameras are NETD and resolution crippled, except the top cameras?
How do you explain, that the Flir One cripple the Lepton 3 sensor from 160x120 to 120x90 (70% area)?
You cant save with the SDK a Flir thermal jpg with 160x120 lepton resolution.

It can form everyone a own opinion.
If I see less details (and more artefacts), then the camera has less NEDT (noise equivalent differential temperature).
less NEDT = more signal noise / disturbances

I also believe, that this is not a bat intent from Flir. But theirs image post processing denoise algorithm removes some details from this consumer camera .
On RAW images of Flir professional cameras like Exx you see the original sensor noise, like from second image below:

here is a real live sample (shot after a small warm up time of about 2 minutes):

It’s the same shot. Compare the different FOV (field of view)

I saved with the SDK.app simultaneously a upscaled Flir Radiometric JPG  and a real Lepton ThermalLinearFlux14BitImage.

Afterwards I rebuild with my old panorama script (see my footer) a real size 160x120 Lepton radiometric jpg (a Flir format).
You can load this sample jpg images in Flir Tools and compare the quality.

First a original image shot with the Flir App.
The App crop    the Lepton sensor to about 120x90 Pixel.
Please note the artefacts/patterns!
Flir makes a nice lens distortion correction of the Lepton sensor for best MSX overlaying 




real  Lepton sensor 160x120 (no image postprocessing and with noise/grain because the temperature spread is only 4 Kelvin)

[Fraser]

@Ben321,

I will not argue with you about noise injection on the F1G2 but please do not call the idea silly as it is proven to be active and present on most of FLIRs thermal cameras ! The E4 had the noise injection wound up high to degrade its NETD compared to other cameras in the series.

Please consider the marketing angle before making statements that noise injection is "silly". FLIR are past masters of thermal image processing but they wish to market different cameras, using similar hardware, at differing price points. Hence why an E4 is just an E8 with its resolution and NETD deliberately degraded. Read the E4 teardown thread for absolute proof of this and the presence of the noise generator on other FLIR camera series, such as the Exx and Txxx.

The FLIR ONE Generation 2 will not be dsigned to directly compete with the more expensive offerings from FLIR, as such is undesirable in their marketing model. Whilst it may be true that the new 12um microbolometer produces enough noise in the F1G2 deployment to satisfy FLIRs needs, it is not "silly" to suggest that noise is added at some stage to ensure market segment appropriate performance, as is the case with so many other FLIR cameras. It is true that the 12um microbolometer finds itself in a far from ideal platform for best performance. The lens size and material is a significant compromise solution that likely reduces performance.

[Ben321]

@Ben321,
I will not argue with you about noise injection on the F1G2

I'm talking about the F1G3. The one that was released for Android phones. I did my testing with an Android phone. F1G2 was only ever available for the iPhone5. I do not have an F1G2. My description that it doesn't appear to have any noise injected, as stated in this thread's opening post, is based on the results of my testing Android-based F1G3 using the sample app that comes with the FLIR One SDK.

The E4 had the noise injection wound up high to degrade its NETD compared to other cameras in the series.

Maybe so, but the FLIR One G3 does not have that. In fact, if you look at the configuration file that comes with the FLIR One SDK, you will see that noise is disabled.

.caps.config.image.targetNoise entry
.caps.config.image.targetNoise.enabled bool false
.caps.config.image.targetNoise.targetNoiseMk int32 0

I assume this is the same configuration file, and from the same SDK, that would have been used by FLIR to compile their official FLIR One app.

Please consider the marketing angle

Actually, if anything, it would hurt marketing. It is nearly inevitable that something like that would be found out, and when it is, it makes the FLIR company look very dishonest (charging thousands more, for something that doesn't have thousands of $ worth of hardware improvements, but just firmware that lets already existing hardware be less crippled, that can easily be modified with changes to a couple lines of code), and that would make their company look bad, and turn away potential customers.

FLIR are past masters of thermal image processing but they wish to market different cameras, using similar hardware, at differing price points. Hence why an E4 is just an E8 with its resolution and NETD deliberately degraded.

But I'm not talking about the E series, or even previous generations of FLIR One. I'm talking about FLIR One gen-3. That's what I was ALWAYS talking about in this thread, right from the opening post. I maybe should have been more specific about that in my opening post, but I didn't think I needed to, because I assumed most people are now using the gen-3. So from now on, all further replies to this thread should be based on the fact that I'm talking about the FLIR One gen-3. The FLIR One gen-2 is irrelevant to the test results I gave describing the performance of F1G3.

Can you prove that the FLIR One gen-3 does in fact inject additional noise? If it really was for marketing purposes, and they were really thoroughly trying to make sure that the end-user truly thought that the FLIR One official app showcased the best possible performance of the device, they'd go out of their way to make sure That the end-user couldn't just program around that. Yet this is not what we see happening. We see that the SDK clearly allows direct access to the 14bit-per-pixel raw data. I tend to think there may be some technical reason why FLIR chose to use cropping in the official app, and why they might have injected any noise (if they even did).

Or are you arguing that even the raw data itself (as is accessible through the SDK) contains noise injected into it?

[Fraser]

Ben321,

No interest in debating this matter with you.

You may wish to note the following so you at least get your terminology correct:

FLIR One = Original FLIR iOS iphone 5 thermal camera using Lepton 2 80 x 60 pixel microbolometer

FLIR One Generation 2 aka FLIR One Gen2 (F1G2) = New FLIR mobile phone thermal camera using Lepton 3 160 x 120 pixel microbolometer. Two varients available, iOS and Android. If you check the FLIR One posts I think you will find that I was the person who started the F1G2 reference so I maybe should have made its meaning clear ?

There is no FLIR One Generation 3 (F1G3). I originally used the term 'F1G2' because I got bored writing it long hand. It is not a reference to the LEPTON Core generation.

Feel free to believe what you like about the FLIR One and FLIR's marketing policy. Its a free world after all 

[mhosier]

Following on from this as a matter of interest (and not debating the noise injection or lack of it), it is not at all unusual for manufacturers to cripple certain products in order to produce a hierarchical product family.  To quote several examples:

1. In the early 90's Intel deliberately disabled the FPU in their 486DX chips, to make a cheaper 486SX version (even though the FPU was on-board and functional and manufacturing costs were the same).

2. A competitor to FLIR for whom I worked some years ago used to make a whole family of pyrometers which were electronically identical and simply labeled them as not having certain features.  99% of customers did not realise (the features were not obvious unless you checked for them) and just bought the top of the range model at nearly twice the price.  Some even sent them back and paid to have them upgraded, resulting in nothing more than shipping the box back to them.  Had the product been sold at the lower price with such a high spec, it would have severely impacted sales of the range above it.

3.  A while back I visited a food plant who made snacks for several retailers.  One retailer sold the SAME product from the same production line at 2 different prices.  The only difference?  The design of the packaging on the cheaper one was designed to look like a value product, so more people bought the "premium" product with the higher profit margin.

It's not silly, it's economics (and very profitable at that).  In the commercial marketplace, most decent sized companies will happily buy a more expensive product with a decent warranty and support, over a cheap product that their staff then have to tinker with in the backroom in order to get decent performance.  This is great news for hobbyists though, who are spending their own hard earned cash and have time to spend playing!

[Ben321]

2. A competitor to FLIR for whom I worked some years ago used to make a whole family of pyrometers which were electronically identical and simply labeled them as not having certain features.  99% of customers did not realise (the features were not obvious unless you checked for them) and just bought the top of the range model at nearly twice the price.  Some even sent them back and paid to have them upgraded, resulting in nothing more than shipping the box back to them.  Had the product been sold at the lower price with such a high spec, it would have severely impacted sales of the range above it.

But they would have had to send the box back, with the firmware modification in place, not just send the box back without doing anything. Otherwise the new features wouldn't be in place.

Also, that seems like a dirty trick. If you are going to sell a product with features that can be added with just a firmware modification, then that means the features are not worth thousands of dollars. This is what I would call an unfair business practice. If they are going to have a cheaper model, it should have actual cheaper hardware in it, so that the more expensive one actually is justified in being more expensive.

[mhosier]

But they would have had to send the box back, with the firmware modification in place, not just send the box back without doing anything. Otherwise the new features wouldn't be in place.

Actually, my whole point was that all the features were already in place and enabled.  They were a range of comms features, and bizzare as it seemed, nobody tended to actually try plugging them in and communicating with them if they hadn't bought the comms version of the device (to be fair, a removable plate covered the connector, so it was not terribly obvious). 

Also, that seems like a dirty trick. If you are going to sell a product with features that can be added with just a firmware modification, then that means the features are not worth thousands of dollars. This is what I would call an unfair business practice. If they are going to have a cheaper model, it should have actual cheaper hardware in it, so that the more expensive one actually is justified in being more expensive.

I can see your point here, and I actually agree with you on the point of principle but the real business case was that it was actually more expensive to make various different circuit boards, populated differently and stocked separately (there were 4 versions, so 4 lots of stock holding costs big $$$) than to make one device in bulk.  So making the 4 separate versions would actually have made the device more expensive all round.  The device cost a few hundred and even if the different versions were created, the hardware cost would have differed by pennies.   

Commercially it is important to have products to fit a customer's needs and pricing.  The top end version had to be priced up so it was not too much cheaper than the range above it, as it was basically "too good" for the money.  But then they also needed it to be an entry level product, so they created a "range" of different versions. I am certainly not agreeing with the principal of it, but economically it was the clear choice.

[Ben321]

But they would have had to send the box back, with the firmware modification in place, not just send the box back without doing anything. Otherwise the new features wouldn't be in place.

Actually, my whole point was that all the features were already in place and enabled.  They were a range of comms features, and bizzare as it seemed, nobody tended to actually try plugging them in and communicating with them if they hadn't bought the comms version of the device (to be fair, a removable plate covered the connector, so it was not terribly obvious). 

Also, that seems like a dirty trick. If you are going to sell a product with features that can be added with just a firmware modification, then that means the features are not worth thousands of dollars. This is what I would call an unfair business practice. If they are going to have a cheaper model, it should have actual cheaper hardware in it, so that the more expensive one actually is justified in being more expensive.

I can see your point here, and I actually agree with you on the point of principle but the real business case was that it was actually more expensive to make various different circuit boards, populated differently and stocked separately (there were 4 versions, so 4 lots of stock holding costs big $$$) than to make one device in bulk.  So making the 4 separate versions would actually have made the device more expensive all round.  The device cost a few hundred and even if the different versions were created, the hardware cost would have differed by pennies.   

Commercially it is important to have products to fit a customer's needs and pricing.  The top end version had to be priced up so it was not too much cheaper than the range above it, as it was basically "too good" for the money.  But then they also needed it to be an entry level product, so they created a "range" of different versions. I am certainly not agreeing with the principal of it, but economically it was the clear choice.

What is the legality then of the E4 mod? If I buy a piece of software that is the "standard" version for $50, but a "pro" version is available for $100, and difference between the standard and pro versions is just a line of code (remove the line of code, and a BUNCH of extra features get unlocked, that actually existed in the cheaper version, but were just locked in the cheaper version by that one line of code), and I hack the software to enable the "pro" features in the "standard" version, the company that sells the software would probably accuse me of "pirating" the more expensive version, simply because I converted the cheaper version into the more expensive version. Doesn't the same principle apply with hardware? If I modify the settings in the firmware of a device, and it enables extra hardware features (such as the E4 to E8 mod), isn't that also illegal?

[mikeselectricstuff]

What is the legality then of the E4 mod? If I buy a piece of software that is the "standard" version for $50, but a "pro" version is available for $100, and difference between the standard and pro versions is just a line of code (remove the line of code, and a BUNCH of extra features get unlocked, that actually existed in the cheaper version, but were just locked in the cheaper version by that one line of code), and I hack the software to enable the "pro" features in the "standard" version, the company that sells the software would probably accuse me of "pirating" the more expensive version, simply because I converted the cheaper version into the more expensive version. Doesn't the same principle apply with hardware? If I modify the settings in the firmware of a device, and it enables extra hardware features (such as the E4 to E8 mod), isn't that also illegal?

ISTR this was done to death in the E4 thread but the bottom line is that the hack, at least the original one, did not involve modifying the software, just the configuration file that told the software what to do.
There might be some tenuous argument that settings are part of the software and subject to copyright but I doubt many people would agree. More so as the settings file is unique for each unit.

[Macbeth]

There is no FLIR One Generation 3 (F1G3).

I'm not sure of that. I've been waiting so bloody long and have already had a deadline extended another month for my Android version I'm beginning to wonder... 

Roll on 4th to 9th May - that's what Amazon are telling me.

[Ben321]

Ben321,

No interest in debating this matter with you.

You may wish to note the following so you at least get your terminology correct:

FLIR One = Original FLIR iOS iphone 5 thermal camera using Lepton 2 80 x 60 pixel microbolometer

FLIR One Generation 2 aka FLIR One Gen2 (F1G2) = New FLIR mobile phone thermal camera using Lepton 3 160 x 120 pixel microbolometer. Two varients available, iOS and Android. If you check the FLIR One posts I think you will find that I was the person who started the F1G2 reference so I maybe should have made its meaning clear ?

There is no FLIR One Generation 3 (F1G3). I originally used the term 'F1G2' because I got bored writing it long hand. It is not a reference to the LEPTON Core generation.

Feel free to believe what you like about the FLIR One and FLIR's marketing policy. Its a free world after all 

I thought that F1G2 meant "Flir One that uses Gen2 Lepton", and that F1G3 meant "Flir One that uses Gen3 Lepton".

[Fraser]

Nope

[encryptededdy]

This whole idea that FLIR intentionally crippled their own product, is pretty silly actually.

While you're probably right in saying that there is no added noise on the F1G2 (in fact you're almost definitely right considering the config file), I don't think the idea that FLIR intentionally crippling their own products is silly, considering what they've done/are doing to much higher end cameras / sensors.

[Ultrapurple]

Just to resurrect an old topic (hey, I'm new here and didn't see it first time round) the practice of selling crippled systems has been going on for decades. This thread https://groups.google.com/forum/m/#!topic/alt.folklore.computers/lZy6zvmxLjc contains reminiscences of old techs who could double disk space with the flick of a switch, remove memory wait-states by changing a jumper and double print speed by removing a relay, not to mention re-enable mainframe features and even multiple CPUs that were 'absent' from the lower-cost models.