HT-18 image noise

[ARD]

Hey guys, new to the forum, but have a quick question for the ones with HTI HT-18 camera.  I have received my unit today and noticed immediately how noisy the image is.  Comparing it to my HT-04 camera with the same sensor it is a night and day difference.  The image on HT-04 is nice and steady whereas the one on the HT-18 is constantly moving.  Take a look at the video I posted on the youtube and let me know what you think.  It looks like I got defective unit, just want to confirm.

[rhb]

Looks like my HT-18.

[ARD]

Interesting,

I took 2 pics of HT-04 comparing to HT-18 and definitely HT-04 images are much better than HT-18 images.  So the HT-18 goes back and I hope the replacement is better.  From what I have heard HT-19 image should be cleaner than that of HT-04.

Here are pics for your opinions.

[Vipitis]

okay, these are great examples. The difference shows. I don't believe this is bad quality control - is this supposed to be the same sensor?

FLIR has been caught to use fake noise. they artificially lower the image quality of their low end products to make the high end look better. An other explanation could be a worse lens because the resolution and sensor seems similar here. Can you compare the lenses form the outside, are they different?

[ARD]

Yes, both sensors are the same between the two and after visual inspection (without opening the cameras) both look identical.  The main difference is that HT04 has 2.4" screen and HT18 has 3.2" screen. Also one needs batteries and the other one is rechargeable, but those two things should have no bearing on the image.   Actually HT18 is considered a higher end camera comparing to HT04

[eKretz]

One thing that would greatly affect the noise level is the temp span. When I set this very low on my Flir E40 (at room temperature the lowest span is about 4°F) it's fairly noisy. Opening up the span even just a bit considerably reduces the noise. In your images there is a different span between the min and max on the different cams, so I'm assuming the total span is different between the two cams. That may be at least part of the issue.

Another worrisome thing I see in your images is the quite large difference in the reported temperatures between the two devices. I assume you took the images pretty close together in terms of time, but there's a 6° - 10°  discrepancy in the temperatures from one camera to the other. That's pretty significant difference at ~10% or more of the reported temp.

[IwuzBornanerd]

One thing that would greatly affect the noise level is the temp span. When I set this very low on my Flir E40 (at room temperature the lowest span is about 4°F) it's fairly noisy.

Yes, narrow spans are definitely revealing!  Could you please post (here or perhaps in the gallery) a couple examples from your E40 at that span (or ~3°C), both gray scale & a color palette?   Also please state the F-stop of the lens.  I realize that "fairly noisy" from a FLIR user's perspective may not be as noisy as what a cheap phone dongle user sees, but I would love to see such an image from an expensive FLIR.  In my mind this is the only way to get a true comparison between cameras.  Except for @Bruno28's revealing images in the gallery from a ThermApp with an F/1.4 lens I can't think of any such low span images posted on this forum (other than mine) although there probably are some.

[bialy100k]

Anybody else can confirm / denied high(er) noise level on HT-18? Sometime this can be just particular unit fault/issue...

[rhb]

The noise pattern on my HT-18 changes about every 3 seconds.

Here are 3 shots with my HT-18

The first is reflection in bathroom mirror with lights off and my hand over the screen.

The second is also a room with the lights off, but no attempt to suppress the screen backlight.  I could not see the guitars in the darkened room.

The 3rd is with the lens cover in place.

I'll let you decide if the noise is objectionable.  If HTI separates the annotation from the image in the JPEG I'll write some code to clean up the row and column noise in post processing.

FLIR has noise suppression pretty heavily covered with patents.  So if the noise in these images is an issue, you'll probably have to buy a FLIR or wait 15 years to buy a thermal camera.

[ARD]

Yes, the temp difference between two cameras was significant.  The images were taken virtually at the same time without touching or moving the object.

As to HT-18 image quality, I think this is isolated issue with my camera.  The refresh rate of 3 sec that rhb sees on his unit is the same as on the HT-04 I have, making the image clearer.  My HT-18 has a constant refresh rate (multiple times per second) making the image appear more grainy.

From everything I have read and pis and video manufacturer send me it looks like HT-18 image should be a bit crispier than that of HT-04.  As soon as I get my replacement unit I will post some pics, still waiting for shipment details.

[eKretz]

One thing that would greatly affect the noise level is the temp span. When I set this very low on my Flir E40 (at room temperature the lowest span is about 4°F) it's fairly noisy.

Yes, narrow spans are definitely revealing!  Could you please post (here or perhaps in the gallery) a couple examples from your E40 at that span (or ~3°C), both gray scale & a color palette?   Also please state the F-stop of the lens.  I realize that "fairly noisy" from a FLIR user's perspective may not be as noisy as what a cheap phone dongle user sees, but I would love to see such an image from an expensive FLIR.  In my mind this is the only way to get a true comparison between cameras.  Except for @Bruno28's revealing images in the gallery from a ThermApp with an F/1.4 lens I can't think of any such low span images posted on this forum (other than mine) although there probably are some.

Sure. I will add them here and some in the gallery. These are in the following order: As received from the camera; after my normal workflow (extract RAW data, upscale/resize, small amount of sharpening/noise reduction) ; with excess noise reduction. I've thrown a few Rainbow palette images in for comparison also. The spans on these are about 5 degrees Fahrenheit for the kid butt and 3.5 degrees Fahrenheit for the miscellaneous desk pictures. According to FLIR the lens for an E40 is F/1.3 IIRC. Of note: excess noise reduction can be as bad as or worse than none at all! Palette can also make a difference in the observed noise level. Rainbow is a bad one, Rainbow HC is even worse.

[Vipitis]

Being in a light or dark room should make no difference. Nowadays the thermal energy emitted by normal light bulbs is low and won't make a difference unless you put your face against it for 2 minutes.

The regular rate that changes noice could be a hit that they do some kind of stacking - do you see any ghosting artifacts when looking at a moving object? Like waving your hand or is every single frame you see completely sharp?

[rhb]

It's doing some sort of recalibration.  There's a shutter on the side of the sensor that opens and closes.

The reason for ensuring total darkness was in case the VL sensor was still mixed in. Also the lights are emitting IR illumination.  Not much perhaps, but the bathroom is all incandescents and the guitar room is a CFL.

I haven't tested it rigorously, but I'm pretty sure the IR sensor is sensitive to VL.

There's a good bit of smear when the scene is changing.

[bialy100k]

if the noise in these images is an issue, you'll probably have to buy a FLIR or wait 15 years to buy a thermal camera.

Already ordered HT-18, waiting for delivery.

[IwuzBornanerd]

Thank you very much @eKretz those are enlightening images, and very low noise, as expected.  To further illustrate the point, and provide comparison, I present a couple shots from my Seek Pro a few minutes after capturing a fresh NUC and no smoothing.  No kid butts here but I do have plenty of tabletop clutter.  The first image is with the grayscale palette autoranged and yielding an 8.9° F range.  With the exception of the uncorrected bias-changed pixels (bright white) this looks to me about the same noise level displayed in the E40 image with 3.5 degree span.  The second image is with the grayscale palette set for 80 shades per degree, yielding a 3.2°F span and you can see much more noise; even more noise than in the rainbow E40 shot.  I'm not sure about the Pro, but the Seek XR aperture is F/1.4.  The aperture opening is the same on the Pro but I have not bothered to calculate the focal length.

I like using grayscale for comparison because I long ago determined that my eyes can only distinguish (at best) 32 shades of gray.  So, for instance, if the palette is spread evenly over a 3.2 degree range I can expect to see differences of 0.1 degree.  I see just a tiny bit of graininess in that E40 shot with the 3.5°F span so I would call it about 60mK (and according to FLIR's site it was spec'd at < .07C sensitivity, so, about right!).  Blowing up & comparing with my Seek shot with 8.9° F range the noise is a bit more pronounced so I would call it over 150mK (but it also is spec'd at "< .07C" sensitivity, so noise exceeds sensitivity).

The whole point of this is that you can't properly compare the noise of one camera with another unless you know enough about the palette & how it is applied to the temperatures.  I took the "rain" and "iron" palettes from @frenky's program and started deleting colors until I could see a difference between adjacent colors.  I reduced the rain palette to 50 colors and the iron palette to 30 colors.  Frenky's rain is similar to the FLIR rainbow palette, so that could explain why @eKretz said the rainbow shows noise worse--because it has more human-recognizable colors than the iron OR grayscale palettes.

So how does this apply to the OP's issue?  If you look at those sample images in "Reply #2", the images from the 04 have a span of around 12 degrees and the ones from the 18 have spans of around 7 & 8 degrees; not quite a 2:1 difference.  A bit hard to say for sure, but I think they are pretty close to the same noise level.

[tmbinc]

What also makes a huge difference is the mapping algorithm. A linear mapping (lowest temperature -> start of palette, highest temperature -> end of palette, with linear interpolation) will produce significantly less noise (thus being better for artistic purposes) than pure histogram equalization (where the goal is to produce an even histogram, i.e. giving temperatures proportional space in the histogram as they have in the image space). 

The latter is better for diagnostic/radiometric purposes, as small hot (or cold) areas won't dominate the range. But a side effect is that when you have an image with large areas of roughly the same temperature, they get allocated a large amount of colors, and noise is amplified significantly.

My understanding is that FLIR uses a combination of linear mapping and hist. equalization, but I haven't looked in close details.

https://www.flir.com/globalassets/imported-assets/document/flir-camera-adjustments-lwir-appnote.pdf has some details on this for Tau cameras.

[bialy100k]

My understanding is that FLIR uses a combination of linear mapping and hist. equalization, but I haven't looked in close details.

Probably best will be to chose (by user) which we want at particular test.

[ARD]

So, I received a replacement HT-18 and noticed 2 things, one good and one bad.  Good thing is the image quality is much, much better than on the previous unit, so that is great.  Bad thing is temperature reading, about 10F or more lower than what I see on HT-04.  So I took out my infrared thermometer and confirmed that temperature reading is much lower.  IR thermometer reads right in line of HT-04, actually two different IR thermometers.
Also it appears that the temp difference or inaccuracy grows with higher temperature, pic of light bulb below - both IR thermometer and HT04 read in 145F range and HT18 in 115F range.  I had the same results when I shot my AC/Heat vent at home.

Not sure what is going on with those, SNs on both HT18s are very close to each other.

I took some pics to compare both cameras.

I have read all of the info in the thread but I'm still at the loss.  Any other suggestions?

What I also notices is that camera pics from HT04 are much nicer and sharper than HT18.

All photos were taken at the same time holding both cameras side by side.

[eKretz]

Eh, something doesn't look right still. I'm kind of confused how there's a color in the image that isn't in the palette range. And that temperature discrepancy is a pretty big problem I'd say. Did you show these pictures to the seller and make inquiry as to what is going on? Looks like they may have a bad run of product or something.

[ARD]

No I did not yet.  Just got this replacement this week and today was the first time I had a chance to play with it.  I will be getting with them on Monday.  I do have to say that the customer service is very responsive.  We will see what they say.

[Lord of nothing]

Hi
Well I am not the only one who is interesting in that devices.
I dont know what the Company is so sorry when I dont use there Name.
I found this:
The sound good for me. I just want see for Home use if some Chips and what ever get hot.
The same for the Floor if the get good heated and if the Insulation works.
I found them on Aliexpress for ~300€ so its good enough for that price?
Or maybe (from that company or other) is better?
Thanks

[rhb]

The reported temperature depends upon the emissivity setting.

Try changing that and see what you get.  Emissivity varies from surface to surface, so one size does not fit all.

Have Fun!
Reg