Research And Application Of Underwater Image Enhancement Algorithm Based On Retinex And Generative Adversarial Network

Images are carriers of information,but also an important tool for humans to explore unknown fields.Ocean is rich in minerals,oil,bioenergy resources and so on,In the process of understanding the ocean,exploring and protecting the ocean,clear underwater images play a crucial role.However,most of the images captured in the real underwater environment will appear blurry,obvious color difference and other problems.Therefore,improving the quality of underwater images is an indispensable step for ocean research.There are two main reasons for the degradation of underwater image quality:first,different wavelengths of light decay at different speeds in water,and the wavelength of red-light decays faster,which makes the underwater image more blue-green.The other is the influence of water and suspended particles on light,which results in forward scattering and backward scattering of light,resulting in low contrast and fuzzy visual effects in imaging.Low quality underwater images can not only bring more effective underwater information,but also cause great trouble to underwater research.In order to improve the reliability of underwater images in underwater research,this thesis mainly focuses on the degradation of the images after underwater imaging.By studying the imaging model of underwater images,thinking about the special properties of light propagation in water,and analyzing various existing image enhancement algorithms,an improved algorithm is proposed to enhance underwater images.At the same time,considering the actual system requirements,an underwater image enhancement system is designed and implemented.The main research content of the thesis includes the following three partsFirst,the halo phenomenon that appears in areas with excessive light and dark after the original image is enhanced with the Retinex algorithm is targeted.In the image preprocessing stage,Retinex algorithm and Gamma correction are used to perform color correction and brightness adjustment on the image,respectively.Then,aiming at the fuzzy details of underwater image,using the dark channel prior algorithm to estimate the transmission map of the image,and the multi-layer perceptron is used for further refinement,finally realizing the enhancement of underwater image.It is obvious from the contrast experiment that this algorithm has better enhancement effect in three aspects of image color,contrast and sharpness than the traditional method and the method proposed by others.In order to achieve more accurate visual perception of underwater images,an improved algorithm for generative adversarial networks is proposed.The loss function of generative network is improved to improve the generation efficiency of generator in the network.Through the adversarial training of generating network(generating image)-discriminating network(judging image),two networks are continuously optimized to enhance underwater images.It can be seen from the comparison that the improved algorithm is better than the underwater generated confrontation network,the enhanced image visual effect is more consistent with human vision,and the problem of color distortion of underwater image is solved.An underwater image enhancement system is designed in this paper.The system consists of three parts:the first part is to preprocess the underwater image using fusion algorithm(Retinex algorithm and Gamma correction);the second part is to enhance the underwater image by using the multi-layer perceptron and the dark channel prior method;The third part is to use the improved generative adversarial network to enhance the underwater image,so that the enhanced image meets the requirements of human vision.