Single underwater image enhancement using integrated variational model

Underwater images often suffer from various degradations such as blurring, fog, low contrast, and color distortion because the light is absorbed and scattered when traveling through water. To solve critical issues, we establish a novel framework combining variational methods and pyramid technology to improve image quality in the frequency domain. Two novel variational models, the adaptive variational contrast enhancement (AVCE) model and the total Laplacian model, are designed with the aim of enhancing the contrast of foreground and preserving texture features at different scales. In order to solve these two models efficiently, we also exploit two optimal algorithms based on gradient descent method (GDM) and alternating direction method of multipliers (ADMM). In addition, fast Fourier transform (FFT) is applied for further accelerating the calculation procedure. Extensive experiments demonstrate that our approach achieves good performance on contrast enhancement, color correction, and texture enlargement for underwater images. Qualitative and quantitative comparisons further validate the superiority of our proposed method. In the quantitative comparisons, the proposed method achieves 1.6170, 0.6484, 0.6333, 0.0332, 4.0355, and 1.6843 scores in terms of underwater image quality measures (UIQM), underwater color image quality evaluation (UCIQE), cumulative probability of blur detection (CPBD), Energy, Entropy, and Contrast metrics, and obtains an average of 10% improvement compared with several state-of-the-art methods. The code is available online at: https://github.com/Hou-Guojia/UIE-IVM.