A Real-Time Two-Point Non-Uniformity Correction Method for Infrared Images Based on Prior Knowledge

The non-uniformity of infrared detectors significantly degrades image quality and detection accuracy, making non-uniformity correction (NUC) a critical research topic. This paper presents an improved real-time two-point correction method that retains the simplicity and high precision of traditional approaches while resolving the issue of frequent recalibration caused by temporal drift. The proposed method vectorizes multi-frame scene images into a concatenated matrix, performs full-pixel sorting along the frame dimension, partitions uniform regions along another dimension, and selects two optimal uniform regions for two-point correction. By employing full-pixel sorting, the method achieves efficient computation of correction coefficients for all pixels in a single operation. Furthermore, the proposed selection method for identifying the two optimal uniform regions leverages prior knowledge from laboratory blackbody images, ensuring better correction results. Experiments results on both simulated and real-world datasets demonstrate superior performance in both subjective and objective evaluations.