Variationally Guided Underwater Image Restoration Using a Noise-Constrained Imaging Model

Light absorption, scattering, color distortion, low contrast and sensor noise are often present in underwater images, resulting in severe degradation and significantly impacting the performance of underwater vision applications. This paper aims at tackling these challenges by proposing a single underwater image restoration scheme by incorporating the underwater imaging model as a guidance for a variational optimization model that is aware of the noise. A region-selection method to estimate the background light is first attempted, which assesses image brightness, gradient information and color deviation. An initial transmission map is then calculated and limited based on the wavelength-dependent attenuation properties for the red, green and blue channels. Then the transmission map is refined to obtain a better depth estimation and visibility recovery of the scene. The variational restoration framework is set up by integrating image priors with a noise model, gradient regularization and color-corrected image priors. The joint estimation of latent clean image and noise components is done using an iterative optimization process. Finally, the BEMD based enhancement and unsharp masking methods are used to enhance the. Experimental results show that the proposed technique is able to successfully restore underwater images while preserving the structural information and noise reduction. For the restored images, good Peak Signal-to-Noise Ratio (PSNR), low Mean Squared Error (MSE), and high correlation coefficients are obtained with the degraded images, which means high visual quality and faithful reconstruction of the scenes. The framework offers a comprehensive solution for underwater imaging applications, including marine exploration and underwater surveillance. The proposed framework is strong to meet the requirements of underwater imaging applications. This paper presents an underwater image restoration algorithm based on variational optimization, wherein the underwater imaging model is used to optimize the contrast of the underwater image. The algorithm comprises of noise suppression, transmission map refinement, background light estimation, BEMD enhancement, image contrast enhancement and underwater vision.