Underwater Image Enhancement Based On Generative Adversarial Networks

The light is scattered and absorbed,resulting in underwater images taken with distorted colors and blurred details.This is detrimental to the exploration of marine life,the protection of marine ecology,and the development of marine engineering.Although underwater image enhancement algorithms have made significant breakthroughs in recent years,improving the effectiveness and robustness of underwater degraded images is still a challenging task.We focus on the technical di culties of underwater image enhancement:(1)Multi-weight and Multi-granularity Fusion of Underwater Image Enhancement.Light is scattered and absorbed as it travels through the water,which results in color shift,poor contrast,uneven illumination,and blurred details.This is not conducive to the exploration of marine life,the protection of marine ecology,and the development of marine engineering.For the problems of visual unnaturalness and blurred details in the enhanced underwater images,this paper proposes a multi-weight and multi-granularity underwater image enhancement algorithm.The algorithm is built on the fusion of two images,which are derived from color-corrected and contrast-adjusted versions of an original degraded image.Based on this,their associated weight maps,i.e.,Laplace contrast weight,local contrast weight,saliency weight,exposure weight,and saturation weight,are normalized and then fused with multi-granularities to solve the problem of the visual unnaturalness of images due to uneven illumination.Further,this paper uses double-scale decomposition to obtain two high-frequency components and then fuse them into image fusion to enhance image contrast and highlight image details.We test the subjective results and objective evaluations of the proposed algorithm on several datasets.The subjective results demonstrate the algorithm not only improves contrast,color naturalness,and brightness but also enhances the details of underwater images.The objective evaluation shows that the average values of UCIQE and PCQI of our algorithm outperform the other six different classical algorithms.(2)Underwater Image Enhancement by Combining Multi-Attention with Recurrent Residual Convolutional U-Net(ACU-Net).From the perspective of deep learning modelbased research,this paper proposed a combined multi-attention mechanism and recurrent residual convolutional U-Net for underwater image enhancementnew roman.fd new roman.fd new roman.fdx(AC-GAN).Firstly,add a dual-attention mechanism and convolution module to the U-Net encoder.It can unequally extract features in different channels and spaces and make the extracted image feature information more accurate.Secondly,add an attention gate module and recurrent residual convolution module to the U-Net decoder.It helps extract features fully and facilitates the recovery of more detailed information when the image is generated.Finally,test the subjective results and objective evaluation of our proposed algorithm on synthetic and real datasets.The experimental results show that the robustness of the algorithm outperforms the other five classical algorithms,such as in enhancing underwater images with different color shifts and turbidity.Moreover,it corrects the color bias and improves the contrast and detailed texture of the images.To address the problems of unnatural image vision,blurred details,and low stability of underwater image enhancement algorithm caused by uneven illumination.We propose a multiweight and multi-granularity underwater image enhancement algorithm and a U-Net network combining multi-attention mechanism and convolution for underwater image enhancement algorithm.It can effectively correct underwater image color bias and highlight image details.Objective results show that the algorithm can enhance underwater images clearly and accurately without generating additional noise and has a certain stability.It is considered to be combined with underwater biological target detection technology to detect underwater resources in the future.