Enhancing Underwater Images with Color Correction and UW-Net Technology

The profound importance of effective underwater image restoration is well-recognized across a variety of domains including underwater exploration, marine biology, environmental monitoring, and autonomous underwater vehicle operations. Current methodologies, however, struggle to adequately address the inherent challenges of underwater imaging, such as severe color distortion and poor visibility due to light absorption and scattering. This paper introduces UW-Net, an innovative end-to-end neural network model specifically tailored for underwater image restoration. UW-Net ingeniously integrates discrete wavelet transform (DWT) and inverse discrete wavelet transform (IDWT) to enhance the extraction and reconstruction of underwater image features. A pivotal aspect of UW-Net is its novel color correction strategy which specifically targets the red and blue spectral distortions, typical of underwater imaging. Additionally, the network incorporates a U-Net architecture with DWT for effective down-sampling and IDWT for precise up-sampling, alongside a chromatic adaptation transform layer aimed at boosting image contrast and color fidelity. Extensive training and evaluation on prominent datasets reveal that UW-Net significantly outperforms existing methods, offering substantial improvements in various performance metrics. This paper delineates the design, implementation, and testing of UW-Net, showcasing its superior capability to restore high-quality visuals from degraded underwater images.