Research On Underwater Image Enhancement Methods Based On Image Formation Model

Underwater archaeology,underwater environment protection,underwater terrain scanning and underwater autonomous navigation,etc.,other underwater activities inseparable from the support of underwater visual technology,obtaining clear underwater images plays a key role in ocean exploration.There are distinguished difference between imaging processing under the water and land.Due to the selective attenuation of light with different wavelengths propagating underwater,the light scattering and artificial light source under the complex underwater environment,etc.,problems,underwater images present blurring,color cast and low visibility,etc.,phenomena.There are two kinds of image enhancement methods for underwater images: image enhancement based on the pixel redistribution and image restoration based on the image formation model(IFM).The former methods focus on transforming the dynamic range of the histogram ignoring the physical model of underwater images,the latter methods have difficulty in correctly estimating the restoration parameters: the background light(BL)and transmission maps(TMs)of red,green and blue(R-G-B)channels.For the purpose of restoring and enhancing diverse types of underwater images,this paper sequentially proposes three effective underwater image enhancement(UIE)methods from multiple perspectives,and comprehensive comparisons of the current mainstream methods indicate the future research direction for UIE methods.The contributions are as follows:(1)For the sake of the simplicity of redistribution-based image enhancement and the effectiveness of the IFM-based image restoration,this paper proposed UIE based on the relative global histogram stretching(RGHS)in different color models.Firstly,the paper pre-processes underwater images based on the theory of Gray-World;and then employs adaptive histogram stretching in the RGB color space according to distribution characteristics of RGB channels and selective attenuation of light propagation under the water;Finally,in order to improve the contrast,saturation and brightness of the image,converting to the CIE-Lab color space,and the brightness and color components are operated as linear and curve adaptive stretching optimization,respectively.Our proposed method avoids the blind enhancement based on the redistribution of pixel values,but improves the visual effect of the image on account of underwater image characteristics and retains available information.(2)In order to obtain the real distance between the scene and the camera,and improve the naturelness of results enhanced by RGHS,IFM-based underwater image restoration methods mainly use prior knowledge to estimate two optical parameters: BL and TMs.The existing BL estimation methods only consider single property of underwater images,and are not applicable for estimating BLs of underwater images under different environments.Dark Channel Prior(DCP)is directly utilized to estimate TMs of underwater images,which often leads to wrong results.In view of the above problems,combining image restoration based on BL fusion and new underwater DCP(NUDCP)with color balancing for UIE is proposed.Improving DCP-based,quadtree-based and MIP-based BL estimation methods and in connection with the influence of brightness on the BL estimation in the whole image,an effective multiple candidate BL fusion method is proposed.NUDCP is proposed through the statistics and analysis of hisgrom distribution characteristics of several high-quality underwater images,which is used to estimate more accurate TMs of RGB channels.In order to equalize the recovered image,the brightness component and (6,(7 color components are operated by standard normalization transformation and optimized adjustment,respectively in the CIE-Lab color model.Experiment results show that improved BL estimation and TM estimation methods can achieve satisfied restoration effects under various types of underwater scenarios.(3)Taking real-time and robust capabilities into consideration,a rapid statistical BL estimation model and an effective TM optimizer are proposed.To obtain reference BLs,a database containing of 500 manually annotated BLs(MABLs)is firstly established,and then considering the correlation between MABLs and the histogram distributions of underwater images,the linear and non-linear BL estimation models of G-B channels and R channel are built,respectively.Because NUDCP-based TM estimation method has a low estimated accuracy for underwater images with the artificial light,the underwater scene depth map based on the underwater light attenuation prior(ULAP),and adjusted reversed saturation map(ARSM)are applied to compensate and correct rough TMs,respectively.Finally,an improved white balance is used to improve the contrast and chroma of the restored image.The experiment results demonstrate that our proposed statistical BL estimation model can estimate the BL more quickly and accurately in contrast with the existing BL estimation methods.The TM optimizer can adjust error regions of TMs estimated based on the NUDCP,and heighten the robustness of the TM estimation.BL and TMs obtained by the proposed method are better than that obtained by state-of-the-art methods in terms of results and efficiency of image restoration.(4)In order to verify the effectiveness of different BL estimation models,TM estimation models and UIE models in the application of underwater images,this paper selects varieties of underwater images with a resolution of 600×400 pixels as experimental data to carry out comprehensive comparisons of UIE methods.First of all,this paper summarizes BL and TM estimation methods and the corresponding prior employed in IFM-based underwater image restoration methods,and then implement objective and subjective evaluation of performance of multiple BL and TM estimation methods and effects of the prior knowledge on the estimated results,finally,compare our proposed three enhancement methods with ten kinds of mainstream methods for recent ten years.Our research results can contribute to understanding advantages and shortcomings of UIE methods and exploring the factors that affect the results of UIE,and provide theoretical support for the low-level visual enhancement of underwater images and have practical application value of auxiliary marine information analysis and mining.