Calibration-Based Polynomial-Fit Non-Uniformity Correction on Thermal Imaging Systems

Abstract

Infrared Focal Plane Arrays (IRFPAs) are used in many areas today as they provide the imaging ability to electro-optical systems. However, the raw images acquired from IRFPAs are usually not useful forthwith since there exist several undesired effects. Non-uniformity (NU) is one of these undesired effects which is caused by several factors such as the cosine-fourth-power effect, photo-detectors on IRFPA being nonidentical, fixed-pattern noise, etc. To eliminate NU and obtain a clearer image, an appropriate nonuniformity correction (NUC) procedure must be applied to the raw image. There are fundamentally two different types of NUC methods that are scene-based and calibration-based. Calibration-based methods rely on adjusting the responsivity of each photo-detector on IRFPA by calibrating the photo-detector response with the help of a uniform radiation source, while scenebased methods rely on correcting the image based on scenery by adjusting NUC parameters in run time with the help of neural-network algorithms, Kalman filter, etc. In this study, a calibrationbased Polynomial-Fit NUC (PFNUC) method is presented and compared to a 2-point NUC (2PNUC) method that is currently used in a serial production thermal imaging system. 2PNUC is one of the most common NUC methods used in thermal imaging systems, where the response of the IRFPA is calibrated based on the raw images taken from a blackbody (BB) at two different temperatures. A gain factor and an offset value are calculated for each photo-detector by using digital signal levels (well-fill level) of the raw image. Ultimately, a gain and an offset table (NUC tables) are formed and applied to the raw image. In serial production systems, each system is usually calibrated at constant BB temperatures. However, this is not appropriate since there may be a considerable amount of variance in the average responsivity of IRFPAs used in systems that may be caused by a couple of reasons such as the different optical transmissions in optics due to imperfection in optical coatings, primitive methods during focus and alignment of the optics, slightly different gain factors on the read-out circuits of the IRFPAs, etc. Given that, when 2PNUC is performed at constant BB temperatures, it leads to both systems being calibrated at different well-fill levels and having different residual NU (RNU) over their dynamic range, which is quite an unwanted characteristic from systems of the same production line. RNU is a measure of how non-uniform is the response of the IRFPA when a uniform radiation source is present in its field of view and an evaluation metric for the NUC method. Despite the average responsivity variance in IRFPAs, the well-fill curve characteristics are quite similar, so it is possible to perform a polynomial fitting. Given that a PFNUC method is proposed and the responses of IRFPAs are calibrated based on a third-order polynomial-fit estimation in which the well-fill levels of each photo-detector are fitted to the median values of IRFPA at four different raw BB images over the dynamic range. By doing so, the coefficients of the fitted polynomial are calculated and the PFNUC tables are formed. A third order PFNUC covers the polynomial characteristics of photo-detectors’ responsivities on IRFPA greatly and it decreases the total RNU by around 50% on average.