Multidimensional fusion of frequency and spatial domain information for enhanced camouflaged object detection

Camouflaged Object Detection (COD) remains a challenging task in computer vision due to the difficulty of distinguishing highly similar targets from complex backgrounds. Existing COD methods often struggle with scene understanding and information utilization, resulting in insufficient accuracy, leading to issues such as background misactivations, target localization losses, and edge blurring. To address these challenges, we propose a novel approach called the Mask and Attention Modulated Information Focusing Network (MAMIFNet), designed to improve COD in the spatial-frequency fusion domain. MAMIFNet introduces a Mask and Attention Modulated Fast Fourier Convolution (MAM-FFC) operator, which adaptively enhances global and local information across spatial and frequency domains through cross-domain masking and attention mechanisms. This operator enables amore comprehensive scene understanding, reducing regional errors such as wide-range false activations and target area losses. To further optimize scene information utilization, the Critical Cue Perception Module (CCPM) is introduced and built upon the MAM-FFC operator. The CCPM refines target foreground focus and foreground-background contrast, mining discriminative cues to improve target localization and edge detection accuracy. Experimental evaluations on four benchmark datasets demonstrate the competitive performance of the proposed method, validating its effectiveness in COD. The code is available at the link below https://github.com/CHANTILLY2023/MAMIFNet.