LEPF-Net: Light Enhancement Pixel Fusion Network for Underwater Image Enhancement

Underwater images often suffer from degradation due to scattering and absorption. With the development of artificial intelligence, fully supervised learning-based models have been widely adopted solve this problem. However, enhancement performance is susceptible quality reference images, which more pronounced in underwater image tasks because ground truths are not available. In paper, we propose a light-enhanced pixel fusion network (LEPF-Net) Specifically, first introduce novel light block (LEB) based on residual (RB) curve (LE-Curve) restore cast color images. The RB learn obtain feature maps an original input image, LE-Curve used renovate learned To realize superb detail repaired superior develop subnetwork (PF-SubNet) that adopts attention mechanism (PAM) eliminate noise image. PAM adapts weight allocation different levels map, leads visibility severely degraded areas. experimental results show proposed LEPF-Net outperforms most existing methods. Furthermore, among five classic no-reference assessment (NRIQA) indicators, enhanced obtained by higher than UIEB dataset.