OK, I have read the thread. Hmmm, did I say I do not like SEEK phone cameras ..... nothing has changed on that front
I think some microbolometer basics are needing to be covered here. Forgive me if you already know all this but it is essential that you understand the basics so I am just making sure.
1. Microbolometer pixels are inherently unstable ..... note this well !
2. Microbolometers using the 12um pixel size appear to be more noisy and less stable that larger pixel Microbolometers. That is just the trade off that must be accepted and dealt with in image processing etc.
3. To use a microbolometer successfully in terms of a decent image there are various issues that must be dealt with....
a) The bias voltages must be set on the microbolometer pixels to provide the required sensitivity. Higher sensitivity is usually accompanied by lower dynamic range. The bias voltages are normally generated 'off chip' by a microprocessor controlled multiple channel DAC. This enables fine tuning of the voltages but they are Global and not pixel specific.
b) the temperature of the microbolometer Die is monitored using a number of diodes on the Die itself. This enables the host system to compensate for Die temperature drift. In a non temperature stabilised Microbolometer, the Die can drift easily and this often limits the sensitivity set using the bias voltages.
Offset tables are used to counter the effects of Die temperature drift but these are often Global offsets and not pixel specific. It is not unusual for geneneric offset tables to be used in budget cameras. This means they are offset tables that are generated from real world testing of many sample Microbolometers to profile the Die Delta T response for the series. That means it is a compromise offset table and not always perfect in every camera. The further away from the 'ideal' Die temperature you go, the more error creeps into the offsets applied. The Die is usually profiled at its nominal operating temperature. In a non stabilised core this is the point at which thermal equilibrium is reached. As a non stabilised system, that Die temperature is subject to change however.
c) A microbolometer FPA is basically a Collection of unmatched thermistors. I say unmatched because they are formed by the same process during fabrication but there is no 'select on test' applied to each pixel ! As such the microbolometer can be a relatively unruly 'beast' that needs taming. Failure to tame the Microbolometers bad habits will result in poor imaging that contains flickering pixels, blotches and temperature gradients across the scene. More detail on taming the Microbolometer later.
d) The scene information coming out of the microbolometer can be very noisy, especially in budget Microbolometers using 12um pixels. This noise is usually dealt with using DSP and produces cleaner images that are pleasing to the eye, yet still accurate in temperature content in a Radiometric camera.
The quality of the video processing often dictates the quality of images produced by any given thermal camera (beyond using decent hardware of course). Poor video processing normal, means ugly pictures.
4. Taming the microbolometer. This is where Ben321 may find something of interest.....
As stated, microbolometer pixels are unstable by nature and prone to output drift. Remember that a thermal scene heats or cools each pixel so it's surface temperature is ever changing. If presented with a constant temperature flat field scene, the pixels come back to a nominal point in their operating range and it is possible to see if any pixels are different to those around them, or outside the acceptable nominal for the majority of pixels.
At the time of manufacture a thermal camera is configured to produce the required picture quality and measurement accuracy. This is commonly called calibration, but in truth it is two processes. Image correction followed by calibration of measurements from the 'corrected' microbolometer ROIC output.
The camera is aimed at a full field thermal scene that is 'flat' with little or no Delta T present. The microbolometer pixels all see the same scene at their surface but there may be differences in the actual energy present on each pixel due to the characteristics of the optics. A computer analyses the output from each pixel and maps the levels ready for correction. Any pixels that are producing a level outside a predefined acceptable range are 'marked' as 'Bad'. Dead pixels are treated as outside the acceptable range so are also captured. Pixels that are providing an output that is within the acceptable level range are further analysed to determine whether any adjustment is needed to the level in order to bring them into the closest possible match across the FPA. There will always be pixels that need to be corrected to make them appear the same as the rest on the FPA. This is Non Uniformity Correction and it can take account of all manner of naughtiness on the part of the microbolometer. The end product is an output to the cameras measuring and display systems that looks nice and 'flat' across the whole FPA. Dead or rejected pixels are disguised using image correction that masks their 'spot'. Such pixels do not adversely effect the measurement accuracy of the camera as they are 'repaired' in the system software.
As already stated, Microbolometers are unruly beasts that have a habit of drifting with time, scene and ambient temperature. The NUC is carried out at a specific ambient temperature and looking at a set scene temperature. It is possible to have multiple NUC correction tables though. These would be created at differing ambient temperatures and different scene temperatures to effectively map the behaviour of the microbolometer in most expected operating conditions. Such techniques are used in shutterless microbolometer cameras. Profiling a microbolometer can be time consuming though, so some manufacturers may choose to cut corners in the way such NUC tables are created.
The NUC process is effective at identifying dead or unacceptable pixels. It generates the base line for each pixels output compensation to achieve a nice flat field across the image. This can also include corrections for a less than perfect lens system. Remember lenses are thicker at the centre than the edges so transmission loss through the lens material does differ across its profile. This is why you can sometimes see a 'cooler' centre to an image when the lens characteristics have not been compensated for.
There is a need for dynamic correction of the output from the pixels as they are anything but stable when in use. As the scene illuminating the FPA is ever changing, it can be hard to keep track of pixels that are drifting away from the others in terms of nominal output level tracking. There are dynamic pixel monitoring algorithms in non shuttered cameras that use scene pixel group analysis to detect such rogue pixels but a shuttered camera uses a known flat field to bring rogue pixels into line.
During a Flat Field Correction event, a shutter or "Flag" is brought into the optical path either between the lens and microbolometer or in front of the lens objective. OEM's decide which they wish to use. The temperature of the shutter/flag is normally that of ambient within the camera or exterior ambient in the case of the external shutter/flag. This temperature and the fact that the shutter/flag is relatively flat in terms of Delta T across its surface, enables a FFC offset table to be produced that results from analysis of all pixels on the FPA. This is a sort of fine tuning pixel gain adjustment map that is dynamic in nature, unlike the factory set NUC tables. If a pixel, or pixels, drift away from the average of the others when looking at a flat field, they are corrected in the FFC table. The global temperature measurement offset may also be extracted from the FFC event in order to maintain camera measurement accuracy during use in different environments. Remember, the FFC shutter/flag has a known temperature. The FFC event has a period dictated by the OEM based upon their knowledge of the Microbolometers stability. Most industrial grade thermal cameras carry out an FFC every 120 seconds or so. They often offer the option to increase this period or even deactivate FFC for video capture. Such deviations from the standard setting normally come with the warning that measurement accuracy and flat field performance will be adversely effected. The SEEK cameras use a very short FFC event period and this reflects the very poor stability of the microbolometer that they are using. It drifts so quickly and badly that the only way to tame it is to have an FFC event every few seconds ! Using a single FFC after warm-up and then running without further FFC events will very likely result in lots of sparkling pixels and blotches on the displayed image within a relatively short period of time. SEEK know this, hence the inability to either increase the FFC period or disablement of it.
Ben321, I do not know the internal software design of the SEEK cameras, but I do know that the OEM was trying to make a Silk Purse out of a Sows Ear with the microbolometer they are using. All manner of effort has gone into taming it as without such it likely produces ugly images that require a lot of processing to tidy up. They already tax the average phone with the current processing required in their Apps. There will be an amount of processing of some sort within the camera so true RAW data may not be available in your case. With that in mind you are effectively 'fighting the system'. You want to do something that the OEM never wanted to happen.... disablement if the FFC event that makes the Microbolometer useable. The 'system' will likely continue to apply FFC offsets no matter what post processing you are doing. As such the offsets may create these troublesome pixels you keep finding.
That is enough from me for now. Please excuse the likely many typos that this response contains as I typed this in a hurry on an iPad.
Fraser