Some trivia on 'Super Resolution' for those interested.
The idea is actually nothing new and dates back to at least the 1970's. It is a legitimate method of increasing the effective resolution of a cameras sensor array.
In early thermal detector arrays there was a need to increase the effective resolution beyond the physical pixel count. In that period, it was common to use scanning type cameras with either one or two scan axis and one or many detector elements, depending upon the design.
In a single axis scanning camera, the thermal scene may be presented to a linear array of detector pixels. If the array consists of only 50 pixels elements, the vertical height of the image produced will be 50 pixels. Now if the detector array is moved vertically such that each pixel fills the gap between the previous scan lines you effectively achieve a 100 pixel height image after two complete scans. The movement of the detector array requires a precision mechanism operating at high speeds. IIRC Piezo Electric drives were used.
Now move forward to the present day and we have cameras stating that they have 160 x 120 pixels yet with a software feature they can produce 320 x 240 pixels. Interpolation is usually the first assumption by many. Whilst a legitimate way to upscale an image, it adds no real new scene data. As such it is flawed in terms of radiometric resolution. Vibrating the FPA sensor at high speed in X and Y planes is a mechanical way to increase the radiometric resolution as the image is effectively scanned to the blind area between pixels. Such a system is complex and expensive though.
It was realised that hand held thermal cameras already had a source of vertical and horizontal scanning..... namely normal human hand shake ! If you imagine the camera capturing a series of images to memory as you hold it in your hand and pointing at a thermal scene, the camera will move slightly between image captures and so when overlayed one on top of the other, there will be image registration errors.The image registration errors are used by the software to effectively generate a higher resolution image that contains real radiometric data samples from the pixels, rather than interpolation.
I have way over simplified the way the software handles the data but this is not just simple image layering as that would likely cause poor image sharpness and blurring. If you mount the camera on a tripod so that it cannot move, the resolution enhancement algorithms have little to work with in terms of camera movement and resultant registration errors in the captured image group. The registration errors are essential to the correct operation of the process.
Before people shout that at 9fps such stacking will severely reduce real time image update rate...... no one said the process needs to operate at the cameras display update rate of <9fps. A microbolometer can be run at 30fps or higher and the images used for Super Resolution can be grabbed out of that high frame rate data stream before the <9fps limitation is imposed on the displayed data. In such a case, the user would notice little or no reduction in frame rate when using the Super Resolution mode. Any decrease would be due to a lack of image processing power in the cameras computer.
So I hope this helps explain Super Resolution to those who were unaware of how it functions. Interpolation it is not
Is it a gimmick ? Well those who used it in the early days of thermal imaging when sensor arrays were seriously low resolution, did not think so. It was a vital and effective means to increase the true pixel data content in an image. Interpolation just 'guesses' the value of a non-existent pixel between others. Good for pretty images, bad for collecting real thermal scene data where every pixel is providing data that is being analysed by the user.
Fraser
hi Fraser
I have been following the thermal imaging treads for a couple of years, and it has definetly
taught me a lot about TIC´s and how they work. I must admit that some of the stuff you and others make theese things do, is over my head, but very interesting to follow.
So i just want to thank you for writing posts like "some trivia on super resolution" it really makes it easy - at least for me - to understand the underlaying things that goes on inside, and in this case how a relative "low" pixel bolometer actually can produce impresive radiometric correct pictures.
I know that you have other things you need to do, but please dont stop making theese very inlightning posts, i think we are a lot of people in here whom really enjoy them.
Have a nice day
Jan