Convexity-Driven Active Contour Model With Heterogeneity–Homogeneity Feature Fusion for Segmentation of Noisy Infrared Images

Infrared imaging plays a critical role in noncontact infrared measurement and detection systems, where a segmentation model serves as a critical preprocessing module within it. However, the accurate extraction of significant measurement targets from infrared images remains challenging due to uneven intensities, edge ambiguities, and strong noise interference. In this article, we propose a novel convexity-driven active contour model (ACM) aimed at enhancing object delineation for detection purposes under noisy infrared conditions. Our method introduces an orientation-selective heterogeneity filter (OSHF) to derive a detailed edge map, which supports the extraction of heterogeneous edge and homogeneous region features. These features are fused via a convex adaptive weighting scheme in the fitting term, while the regularization term is guided by edge-aware information. Optimization is achieved using the Split Bregman method, enabling globally convergent energy minimization. The experimental results on noisy infrared datasets from real infrared monitoring scenarios demonstrate significant improvements over traditional ACMs in terms of segmentation accuracy and robustness. The proposed model thus enhances the reliability of infrared systems by enabling more precise localization and quantification of target regions.