Least Square Regression based Non-Uniformity Correction for Infra-red Focal Plane Arrays

In last few decades, a considerable amount of research has been done in the development of Infra Red (IR) Focal Plane Array (FPA) technology. Due to this, there has been a proliferation of higher sensitivity and larger format FPAs with smaller pixel pitch. Even with the state-of-art VLSI manufacturing techniques, variations in response of individual detector elements of FPA results in spatial non-uniformity for isothermal blackbody source. Generally, two-point Non Uniformity Correction (NUC) technique is used to tackle this problem. For reducing Residual Non Uniformity (RNU) the entire operating range of sensor is divided in segments and piecewise two-point NUC is performed. The offset and gain coefficients of each such segment are calculated and stored in the form of table in non-volatile memory. During the image formation process these coefficients compensate spatial non-uniformity. In the present approach a linear Least Square Regression based NUC (LSR-NUC) technique is suggested, which relies on knowledge of the true irradiance and corresponding detector outputs at distinct temperature levels. A close approximation of irradiances at these points is explored in the form of the best fit line by minimizing sum of square of errors. In this way a gain and offset value for each detecting element is generated and recorded in memory, which is utilized during real time image formation process. After performing LSR-NUC, RNU of FPA is calculated and its comparison is done with RNU calculated after conventional two-point NUC and a remarkable gain in performance has been observed specially in temperature ranges which are not in the vicinity to points chosen for two-point NUC.