基于全局背景光强度估计和对比度调整的无监督水下图像增强

An unsupervised underwater image enhancement method with global background light intensity estimation and contrast adjustment is proposed. Aiming at the inaccuracy of Gaussian blur in estimating global background light，a novel method is proposed. It utilizes a network incorporating upsampling，downsampling，and convolution to estimate the global background light，with a smoothness constraint to supervise the network’s learning. To tackle the issues of color cast and low contrast caused by complex imaging environments and imbalanced local‐global processing，a comprehensive approach for contrast adjustment is introduced to impose contrast constraints on the network. In addition，this approach integrates automatic levels and the contrast limited adaptive histogram equalization（CLAHE）method to enhance global brightness and contrast while preserving local details. Experiments were conducted based on a self‐constructed underwater image dataset and a real underwater image dataset. The results indicate that this method achieves outstanding performance in visual effects and quantitative indicators. The images in the self‐constructed dataset exhibit high contrast and clear details after enhancement，outperforming others in terms of underwater color image quality evaluation（UCIQE），underwater image quality measure（UIQM），and a combined metric，achieving scores of 0. 54，0. 55，and 44. 76，respectively.