QFFS‐Net: Quality‐Guided Fuzzy Sigmoid Selection Network for Low‐Light Image Enhancement

Low‐light image enhancement remains a challenging problem due to poor illumination, increased noise, and color degradation. In this paper, we introduce a new enhancement framework based on quality‐guided fuzzy sigmoid selection network (QFFS‐Net). First, multiple enhancement candidates are generated by applying fuzzy‐based sigmoid transformations to the luminance channel in the Lab color space. These candidates are then evaluated by a lightweight convolutional neural network that assigns each a perceptual quality score. The predicted quality scores are normalized using a softmax‐based fusion strategy to combine the enhancement candidates into a fused intermediate image. This intermediate image is subsequently processed by a lightweight residual refinement network to correct residual errors and produce the final enhanced output. This design allows the model to exploit complementary information from multiple enhancement hypotheses instead of relying solely on a single transformation. The proposed method achieves comparable results of PSNR and SSIM against the benchmark datasets LOL and LOLv2‐Synthetic. It also provides visually consistent results on no‐reference datasets such as LIME and DICM. The proposed method provides a better trade‐off between enhancing lighting, maintaining structure, and color fidelity compared to existing methods. QFFS‐Net offers an effective and computationally efficient approach to low‐light image enhancement without relying on complex end‐to‐end architectures or large training datasets. The work code can be downloaded via the link: https://github.com/hazimghd/QFFS-Net .