CSAR-Based Nonuniformity Correction with Adaptive Ghosting Suppression via a Multi-Mechanism Approach

Infrared focal plane arrays (IRFPA) suffer from fixed pattern noise (FPN) arising from pixel-to-pixel response variations, which manifest as bias and gain nonuniformities, with gain nonuniformity dominating the linear response model. The existing constant statistics of adjacent ratios (CSAR) method can effectively correct gain nonuniformity, but is prone to ghosting artifacts in static or slow-moving scenes, and suffers from high storage requirements and slow convergence. To address these issues, this paper proposes a nonuniformity correction method based on CSAR with adaptive ghosting suppression, termed CSAR-AGS. Building upon the gain-correction advantage of CSAR, the proposed method employs a cooperative mechanism of flat region identification and motion detection, updating the ratio statistics only under conditions of global frame motion and spatially flat regions. An adaptive convergence criterion is introduced to divide the algorithm into initial correction and ghosting suppression stages, effectively avoiding ghosting artifacts. Meanwhile, recursive estimation of the arithmetic mean replaces median filtering, substantially reducing storage and computational overhead. Theoretical analysis proves the convergence of the algorithm and establishes an error propagation model. Experimental results demonstrate that CSAR-AGS effectively corrects gain nonuniformity and, under the condition of sufficiently low bias-to-signal ratio, can be extended to bias and mixed nonuniformities, outperforming existing methods in convergence speed, correction accuracy, and ghosting suppression. The algorithm converges within several hundred frames, with a per-frame processing time not exceeding 0.0110 seconds, showing potential for real-time deployment.