Research on Underwater Image Enhancement Technology Based on Physical Modeling and Deep Learning

The complexity of the underwater environment leads to three main degradation problems, namely, color deviation, low light and blurring, as the image quality is affected by factors such as light absorption and scattering. Color deviation is caused by light absorption and scattering, which is manifested as color distortion; low light is caused by light attenuation, which results in insufficient brightness; and blurring is caused by scattering of water particles, which blurs the edge information. In this paper, based on the physical model of underwater images, the degradation phenomenon is decomposed into direct, forward scattering and backward scattering components, which correspond to color deviation, blurring and contrast degradation, respectively, and evaluated based on PSNR, UCIQE and UIQM metrics. Three enhancement methods are proposed for single degradation problems: MSR technique to improve light unevenness, CLAHE to enhance brightness and details, and Laplace operator to sharpen edges to enhance clarity; for complex degradation scenarios, combining wavelet transform and GAN model, wavelet decomposition is used to preserve the details of the image, and GAN to optimize the realism and quality of the image. Experimental validation shows that the above methods are excellent in optimizing underwater image quality, and have good adaptability and practical guidance.