Underwater Variable Zoom: Depth-Guided Perception Network for Underwater Image Enhancement

TL;DR

A novel depth-guided perception UIE framework, dubbed underwater variable zoom (UVZ), which parses near-far scenarios by harnessing the predicted depth maps, enabling local and non-local perceiving in different regions.

Abstract

Underwater scenes intrinsically involve degradation problems owing to heterogeneous ocean elements. Prevailing underwater image enhancement (UIE) methods stick to straightforward feature modeling to learn the mapping function, which leads to limited vision gain as it lacks more explicit physical cues (e.g., depth). In this work, we investigate injecting the depth prior into the deep UIE model for more precise scene enhancement capability. To this end, we present a novel depth-guided perception UIE framework, dubbed underwater variable zoom (UVZ). Specifically, UVZ resorts to a two-stage pipeline. First, a depth estimation network is designed to generate critical depth maps, combined with an auxiliary supervision network introduced to suppress estimation differences during training. Second, UVZ parses near-far scenarios by harnessing the predicted depth maps, enabling local and non-local perceiving in different regions. Extensive experiments on five benchmark datasets demonstrate that UVZ achieves superior visual gain and delivers promising quantitative metrics. Besides, UVZ is confirmed to exhibit good generalization in some visual tasks, especially in unusual lighting conditions. The code, models and results are available at: https://github.com/WindySprint/UVZ.

Figures (17)

• Figure 1: Near-far scenarios enhancement comparison by different methods, where the orange box represents the enlarged near scene and the red box represents the enlarged far scene. The top row shows the raw image, the results of L$^2$UWE 5, CBM 6, TEBCF 7, WWPE 8, and Water-Net 9. The bottom row shows the results of SCNet 10, TACL 11, USUIR 12, PUGAN 13, Spectroformer 14, and our UVZ.
• Figure 2: Motivation of our proposed depth-guided perception. The degradation degree varies significantly between the near scene and far scene. In comparison, the visual quality of the former and the adjacent regions exhibits relatively better performance. For the two scenes, we should implement different feature learning strategies.
• Figure 3: The two-stage framework of UVZ, including DEN, DGEN, and ASN. All sub-networks adopt a standard encoder/decoder architecture, where the red slash indicates that ASN is only used for the training. In the first stage, for a raw image $X$, DEN and ASN generate the depth map $d$ and the regression image $\widehat{X}$, respectively. For the second stage, with inputs $X$ and $d$, DGEN generates the enhanced image $Y$.
• Figure 4: R$^3$S transformation of a depth map. The input is the depth map d and the outputs are the feature maps $d_{rev}$, $d_1$, $d_3$, and $d_5$. The process within the dashed line is a 3 × 3 region smoothing (RS).
• Figure 5: Structure of the Dual-Attention Module (DAM)