Blind Restoration Of Underwater Images Based On A Complete Optical Imaging Model

In an underwater environment different from the atmosphere,the density of the water body is much greater than that of the atmosphere,resulting in more suspended particles and stronger absorption and scattering of light.Thus underwater captured images often suffer from haze,low contrast,noise and cast,which make it difficult for application.Single underwater image often has multiple quality degradation problems simultaneously,so improving the quality of the degraded image and recovering a clear image is of significance to the practical application of the image and scientific research.Most of the existing restoration and enhancement methods are impossible to effectively repair every degradation problem.Aiming at the problems of defogging,denoising and deblurring of underwater images,researchs on blind restoration of underwater images are as follow:Underwater images are usually degraded which limits their analysis and application.To address these problems,a red channel prior guided variational framework is proposed based on the complete underwater image formation model(UIFM).Plenty of the existing methods only consider the direct transmission and backscattering components,we additionally include forward scattering component into the model.In the variational framework,we incorporate the normalized total variation item and sparse prior knowledge of blur kernel together.For solving the generated non-smooth optimization problem,we employ the alternating direction method of multipliers(ADMM)to improve the convergent speed.And we perform the estimation of blur kernel by the image pyramid in a coarse-to-fine manner to avoid local minima.Extensive qualitative and quantitative comparisons further validate its superiority against the other state-of-the-art algorithms.A novel variational restored approach based on the sparse prior of red channel is proposed according to complete the UIFM.To simultaneously overcome haze,low contrast,cast and blur,multiple regular terms with different purposes are introduced into the variational energy equation.First,the data fidelity term of the variational equation is designed,based on the forward component of the UIFM.Subsequently,L0 norm of image red channel serve as the deblurring regular term and the prior that the red channel of clear underwater image is sparser derives from numerous observation.Finally,pre-restored image is introduced as guidance for color recovery.To further accelerate the algorithm efficiency,the ADMM is used to solve proposed model.Experimental results indicate that proposed method not only achieve dehazing and denoising,improve contrast,recovery color,but also removal blurry and optimize clarity.