Research On Underwater Image Enhancement Based On Multi-scale Fusion

Underwater image is an important carrier of underwater information that has been widely used in underwater robots,marine archaeology,and marine energy exploration etc.However,due to the special physical properties in that environment,underwater images suffer from serious degraded problems such as blurred details,color distortion and low contrast,which affect the extraction of image information.This thesis proposes an underwater image enhancement method based on multi-scale fusion,comprehensively considering the visual effects of images,the perfection of image information and the complexity of the algorithm.The main work of this thesis is as follows:Firstly,we establish the underwater imaging model of attenuation and scattering by analyzing the optical properites of water,and describe the main features of the underwater image.The current major underwater image enhancement methods are studied and implemented,including histogram equalization,white balance,Retinex enhancement,dark channel prior and image filtering.Secondly,aiming at two main image degradation problems of color distortion and contrast reduction,we design an effective underwater image enhancement method based on multi-scale fusion.The improved white balance,histogram equalization based on Lab color space and gradient domain guided filtering methods are used to obtain the color corrected and contrast enhanced versions of original degraded image as the fusion inputs.The normalized weight maps are calculated by selecting four weighting features such as saliency map and chromaticity diagram.The final result is obtained by multi-scale decomposition and fusion of inputs and weight maps using Laplacian pyramid and Gaussian pyramid with high efficiency.Finally,this thesis establishes a set of underwater image datasets with various degradation characteristics,and evaluates the multi-scale image fusion method by utilizating the current excellent non-reference underwater image evaluation methods.The effectiveness,adaptability and stability of the proposed algorithm are illustrated from two aspects: visual effects and image quality evaluation criteria.