NUIENet: A Novel End-to-End Underwater Non-Uniform Illumination Image Enhancement Method

The main innovative achievements of this paper are as follows:

(1) Addressing the problem of non-uniform underwater image illumination caused by artificial light sources, a method called NUIENet is proposed for underwater non-uniform illumination image enhancement. NUIENet extracts and fuses multidimensional semantic features such as brightness, chromaticity, and saturation by utilizing a parallel correction network and fusion layer in an end-to-end manner. Initially, an encoder-decoder structure with skip connections is employed to parallelly correct the image channel-wise, achieving synchronous enhancement of different semantic features. Then, a fusion layer is designed to optimize the spatial representation between different channels, further improving the quality of the output image. In addition, to address the lack of an effective training database, a method based on generative adversarial networks and Gaussian functions is devised for synthesizing underwater non-uniform illumination images by analyzing the propagation characteristics of light underwater and the characteristics of artificial light sources. Comparative analysis with various state-of-the-art methods on multiple databases like NUII and OceanDark demonstrates the superiority and robustness of NUIENet in improving image illumination distribution, color balance, and contrast.