A Review of Gan-Based Texture Reconstruction of Underwater Images

With the deepening of marine resources development, the importance of underwater image processing technology is becoming more and more prominent. Nevertheless, such images frequently exhibit colour distortion, diminished contrast, and blurred textures due to light scattering and absorption. Although traditional image enhancement methods are effective, they have limitations such as noise amplification and poor environmental adaptability. In recent years, methods based on the Generative Adversarial Network (GAN) have shown significant advantages in texture reconstruction and colour restoration by learning underwater image features through adversarial training. The paper systematically reviews GAN-based texture reconstruction methods for underwater images, comparatively analyze the performance differences of multiple models from early to present and test them on underwater image datasets to quantitatively evaluate their effectiveness based on image quality indicators. The experiments have demonstrated that the method based on Generative Adversarial Network (GAN) outperforms traditional approaches in terms of detail restoration and generalization ability. However, it still has problems such as high computational complexity and data dependence. Future research can combine physical modelling and lightweight design to further enhance real-time processing capabilities and environmental adaptability.