UNDERWATER IMAGE ENHANCEMENT TECHNIQUE BASED ON CYCLEGAN AND FREQUENCY DECOMPOSITION CORRECTION MODEL

Enhancing underwater images is a challenging task due to degradation from light scattering, absorption, and low contrast, which obscure important details. This paper proposes a novel method that combines Cycle-Consistent Generative Adversarial Network (CycleGAN) with Frequency Decomposition Correction to improve underwater image quality. Initially, CycleGAN was used to generate clearer images from blurred underwater photos, addressing distortions caused by the underwater environment. Next, Frequency Decomposition Correction separates the image into low-frequency (smooth areas) and high-frequency (details and edges) components. The low-frequency component is enhanced for better clarity, while the high-frequency component is sharpened to improve fine details and edges. The enhanced images are evaluated using PSNR, UCIQE, and UIQM metrics. Results show that CycleGAN significantly improves image quality, and the additional Frequency Decomposition Correction boosts clarity, contrast, and detail further. This combined method presents a promising solution for underwater image enhancement, with potential applications in marine research, underwater navigation, and environmental monitoring.