Corrección de imágenes IR mediante un filtro extendido de estadísticas constants

Abstract

The quality of an infrared focal plane array camera (IRFPA) in terms of image sharpness is conditioned on the sophisticated manufacturing process of its sensor stage. That is, it is hard to build photo detectors with exactly the same response of electrical signal. This problem is known as “non-uniformity” in IR technology and it manifests itself as superimposed grid in the output image of the camera, termed as fixed pattern noise (FPN). This noise emerges since the weak electric signals from the detectors must undergo a high gain amplifier stage, thus magnifying their differences notoriously at the exit of the camera. To address this problem, the detector is characterized as a linear model with two parameters (gain and offset). To find these parameters and counteract this inequality we propose an algorithm based on a nonlinear digital filter extended from a simple yet consistent experimentally verified theoretical development of the standard method of Constant Statistics (CS). We demonstrate that the new filter compares favorably with CS standard, in terms of convergence speed and therefore prompt fading of ghosting artifact or “ghosting”. Parameters of the proposed algorithm were adjusted and when it was tested with synthesized and real infrared video, high levels of correction was achieved, notoriously decreasing the non-uniformity.