Multi-focus image fusion (MFIF) aims to extract focused regions from images captured at different focal depths and generate a single all-in-focus image to enhance visual quality and information utilization, thereby broadening its application scope. To address the limitations of existing methods in detail preservation, complementary feature modeling, and structural information retention, we propose a novel MFIF method named CFANet. Our approach introduces a frequency-domain attention mechanism based on the Fourier transform to fully exploit multi-frequency features within the image and enhance the preservation of structural information. Furthermore, we design a cross-attention mechanism for low- and high-frequency components to enable effective feature interaction and enhancement between source images during the feature reconstruction phase. This design alleviates the common issue of relying excessively on single-image features while neglecting global consistency. Extensive experimental results demonstrate that CFANet outperforms existing methods across multiple mainstream evaluation metrics, particularly excelling in detail sharpness and structural consistency. The code for this work is publicly available at: https://github.com/lasiaYen/CFANet .