Camouflaged object detection is a challenging task in computer vision, focusing on accurately detecting and segmenting camouflaged objects in images. Despite the relatively high accuracy of existing models, challenges remain in improving localization, segmentation precision, and reducing model parameters and computational complexity. To address these critical issues, we propose GFNet, a camouflaged object detection network with Global cross-space perception and Flexible local feature refinement (GFNet). GFNet effectively integrates global perception and local feature refinement to achieve more precise detection and segmentation. It introduces the Multi-scale Cross-space Global Perceiver (MCGP) for multi-scale global localization and the Feature-guided Refinement Decoder (FRD), which extracts and integrates contextual information through a novel Partially Interactive Deformable Convolution (PIDConv). This enables the model to reduce computational cost while enhancing segmentation accuracy. Experimental results demonstrate that GFNet outperforms 14 state-of-the-art methods across four commonly used datasets, achieving superior performance in detection accuracy, model size, and overall computational efficiency. The code will be publicly available at https://github.com/ForestFireBoy/GFNet .