<p>Multi-exposure image fusion (MEF) aims to combine multiple low dynamic range (LDR) images into a single high dynamic range (HDR) image with enhanced details and balanced exposure. While deep neural networks (DNNs) have driven substantial progress in MEF, challenges remain in producing results with rich details and vivid colors. To tackle these issues, we propose a perceptual enhancement network that incorporates supervised learning in both the spatial and frequency domains. Specifically, an attention fusion module is employed to extract key information from source images for preliminary fusion. This is followed by a collaborative correction module, which enhances fine details and corrects color distortions in the fused image, resulting in visually pleasing outputs. In addition, we introduce a frequency-domain supervision loss based on the fourier transform to measure the discrepancy between the fused image and the ground truth. This guides the network to better capture image structures and helps alleviate spectral bias. To further improve perceptual quality, we adopt an adversarial training scheme involving both spatial and frequency domain discriminators. Extensive qualitative and quantitative experiments on diverse datasets demonstrate the strong capability of the proposed method in multi-exposure image fusion.</p>