Research On Super-resolution Underwater Image Enhancement Algorithm Based On Adaptive Color Compensation

The unpredictability of the underwater environment and multiple disturbing factors lead to color distortion,low visibility and blurred details in underwater images,which seriously affect the application of vision-based techniques.In order to extract useful information from the scene,image enhancement and restoration are needed to improve the image quality.Therefore,in this thesis,with the research objectives of enhancing the visibility of underwater scenes,eliminating image color bias and improving detail blurring,we synthesize underwater image datasets by combining underwater optical imaging mechanisms,and also conduct research on underwater image enhancement and super-resolution.The main work is as follows:(1)For the problem of insufficient underwater image data,an underwater image generation method based on the imaging model is proposed.The underwater imaging model is used as the main body,and a dual background light adaptive fusion method is designed to improve it.A monocular depth estimation algorithm and a weighted bootstrap filter are introduced to obtain a smooth depth map of the natural image,and then the transmittance is calculated.The natural image is degraded to an underwater image by combining the requested relevant parameters.The attenuation coefficients of different water quality types are selected,and the underwater images corresponding to the water quality types can be synthesized.(2)An underwater image enhancement algorithm based on multiscale color compensation and fusion is proposed for underwater images facing color distortion problems.A color shift detection algorithm is introduced to classify the images.A multiscale color compensation method is designed to compensate the channels for color-biased images,and then white balance the compensated images.A rank-one prior matrix-based image defogging algorithm is used to implement underwater image defogging,and an improved unsharp mask method is used for detail preservation.Finally,the white balanced image is fused with the sharpened image in CIEL*a*b*space to obtain the final enhanced image.Comparative experiments were conducted with eight classical algorithms on the public and synthetic datasets.From the qualitative and quantitative analysis of the experimental results,it is clear that the proposed method in this chapter can effectively eliminate the image color bias,improve the image contrast,and have better robustness.(3)An underwater image enhancement method based on adaptive color compensation and super-resolution reconstruction is proposed for underwater images that face the problems of both color distortion and low resolution.The adaptive color compensation method is designed to pixel-complement the color-distorted images by combining the underwater attenuation characteristics.A gamma correction method is used to suppress and enhance the over-bright and over-dark areas of the image.An underwater depth-coupled feedback network is designed to fully fuse the above two images and improve the image resolution by using the mutual collaboration between the modules.Also,a hybrid loss function supervised network training is designed.Adequate comparison experiments were conducted on the public and synthetic datasets,and the results show that the method further solves the problem of low image resolution based on underwater image enhancement,resulting in more realistic image tones and clearer textures.