Manufacturing Trend 2016/06, Technical Diagnostics Section

"Accurate temperature measurement or impressive visualization?"

IMAGE ENHANCEMENT IN THERMAL CAMERAS (I.)

What do we need for precise thermographic measurement, what do manufacturers/distributors promise, and ultimately what do we get from them?

The background of our new series is an article published in our magazine last year under the title "Modern thermal cameras from a professional perspective," focusing mainly on the possibilities of enhancing images obtained with thermal cameras. The expanded version of these sections, as well as their continuation, can be read on the Manufacturing Trend pages starting from this issue. The reason for revisiting the topic is partly due to the emergence of new methods and procedures in image enhancement since the previous overview. Another motivation for the continuation is the unfortunate experience that some market players offer seemingly equivalent technologies with correct measurement procedures at a favorable price. However, the end user without the appropriate knowledge remains unaware that the measurement capability of the given thermal camera does not even come close to the promises. In other words, neither the pixel resolution is higher (not even as much as stated in the datasheet), nor is the thermal image more accurate or sharper. Instead of accurate results, at best, they receive enhanced graphical representation but worse measurement data! Thus, what seemed cheaper quickly becomes very expensive. The image quality achievable with a thermal camera, or more precisely the detail of the measurement, is determined not only by the geometric resolution but also by the number of image points detected within the field of view of the thermal camera. The reason for this is that for graphical (visual) recognizability, a certain minimum number of image points must fall on certain parts of the object being measured – just as we are used to in digital photography. It is easy to understand that with more image points, we can represent the object surface with greater detail or the same level of detail over a larger object area in a thermal image. If the number of points is insufficient to create a thermal image of the expected quality, many partial images of the object need to be taken, and it may become necessary to montage the images, which is a very time-consuming task. First, we present two software-based thermal image pixel-increasing procedures carried out within the thermal camera's field of view to eliminate this, both characterized by expanding the data stored in the measuring device and the data file. Resolution enhancement by interpolation Due to the relatively small number of pixels in thermal camera sensors, creating impressive thermal images (and consequently reports) poses great challenges – especially with thermal cameras having a smaller number of pixels in their sensor matrix. To alleviate this problem, some thermal camera manufacturers apply interpolation commonly used in graphic image processing programs. The procedure generates an additional – mathematically interpolated – image point between each pair of pixels in the captured thermal image. This increases the pixel count of the thermal image fourfold by doubling it horizontally and vertically. However, this procedure results in a thermal image that is 75% calculated, thus not containing real, measured image points. Therefore, improving the visual appearance of the thermal image comes at the cost of distorting the data content of the image. From a metrological perspective, the application of this procedure is not recommended.

On the datasheets of some companies' so-called "low-cost" thermal cameras, which are marketed as professional but are actually at a low-cost level, the thermal image pixel resolution is given in the following form, or with the text "320×240 pixels (interpolated)" or possibly "320×240 pixels (interpolated)," actually representing a detector of 160×120 pixels. The data file of this is filled with 75% false data through interpolation to make them appear competitive with thermal cameras containing a real 320×240 pixel matrix detector. Neither the real image resolution, nor the geometric resolution, nor the content of the data file can be compared with those of a detector-based thermal camera with the latter number of pixels. It is unacceptable to fill the measurement file with interpolated data because it is impossible to determine afterwards whether it is real measurement data or inserted values. Consequently, correct evaluation is already excluded. The situation is illustrated in a thermographic image of a heating fan.

In the image on the left in Figure 3, due to pixel deficiency and insufficient geometric resolution, a false measurement results in only 363 °C being displayed as the maximum temperature, although the actual temperature exceeds 425 °C. The thermal image produced with interpolation on the right is 320×240 pixels in size, but beyond its visually enhanced appearance through pixel increase, no more accurate or correct measurement results are obtained. Interpolation did not improve the lack of measurement pixels or increase the geometric resolution. This is evidenced by the fact that the displayed maximum temperature value remains only 363 °C.

In the image on the left in Figure 4, the recording of the same heating coil with a 320×240 pixel detector is shown. Thanks to the larger number of pixels and the increased geometric resolution under the same optical conditions, not only did the quality of the visual appearance improve, but the measurement result also became more accurate. This is supported by the fact that the displayed peak value of 412 °C now more closely approximates the actual maximum temperature. The best image is achieved with a 640×480 pixel detector since this pixel count (under the same optical parameters) meets the measurement requirements for geometric resolution. Thanks to correct measurement, the maximum of 425 °C was displayed, and the visual appearance became more valuable.

Resolution enhancement utilizing hand tremor

In reality, a sensor matrix consists not of individual sensors placed seamlessly next to each other, but there is an (almost half-pixel) insensitive, non-measuring gap around each sensor. Therefore, the detection of the measured object also occurs only "sparsely". In order to eliminate this, in recent years, a different software-based method for increasing thermal image pixel resolution has started to become popular instead of interpolation (e.g., under the names Super Resolution or UltraMax). These methods are based on the slight horizontal and vertical field-of-view shifts caused by the hand tremor or movement of the person holding the thermal camera.

The method is quite simple: instead of one thermal image, the data of (typically) 16 thermal images are stored, and then, with the help of software, we select four images that "fit" together with a half-pixel horizontal and vertical shift due to hand tremor. Then, we align the thermal images pixel by pixel next to each other or below each other. With this method, data is obtained even from the empty space between the originally two adjacent elementary sensors (pixels). The number of pixels is doubled horizontally and vertically - our thermal image will have four times the resolution compared to the original detector matrix pixel count. Moreover, since the detection of the (same-sized) field of view is now seamless, the geometric resolution of the thermal camera also improves by exactly 33 percent. (In advertising brochures, other numbers are often found, but these cannot be supported professionally.)

As simple (and inexpensive) as this method is, it also comes with many pitfalls. In the case of a tripod-mounted thermal camera, it cannot be used at all, and the hand tremor of a person is only rarely regular enough for the software to find four images among the stored 16 thermal images that can be aligned with each other as described earlier. (Just think about the fact that the whole process takes nearly half to one second. If our hand tilts or continuously moves during this time, the four alignable thermal images will not be obtained in any way.) Furthermore, the software image selector algorithm is also unable to select thermal images in every case where the thermal image does not contain sufficiently large and sharp contrasts (adequately steep temperature gradients), or if there is a displacement in some part of the field of view.

Rahne Eric (PIM Ltd.) pim-kft.hu, termokamera.hu

Contact

The content of the publication is protected by copyright. Any (even partial) use, electronic or printed re-publication is only permitted with the indication of the source and the author's name, as well as with the author's prior written permission. Violation of copyright (Copyright) will result in legal consequences.