Research On Underwater Optical Image Style Transfer And Color Correction Algorithm

The ocean is rich in information,and is important for the sustainable development of mankind.Obtaining valuable information from the ocean through images is a necessary means in the process of exploring the mysteries of the ocean.In practical applications,submarine terrain scanning,submarine communication cable laying,underwater vehicles and underwater search and rescue are all inseparable from the development of underwater image processing.However,in the process of acquiring underwater images,there are challenges from various as-pects.Firstly,when light travels through the water,the water will selectively absorb some of the light and cause the color of underwater images to deteriorate.Due to the influence of floating substances in the water,th e light will have a certain degree of scattering,which brings problems such as blurred details and low contrest to underwater images.In addition,due to the limita-tions of hardware devices,there are difficulties in obtaining high quality underwater images.Therefore,using image processing technology to restore the original appearance of underwater images plays an important role both in marine science research and industrial applications.This paper focuses on underwater image processing technology,exploring methods for underwater image style transfer and color correction.There are many reasons lead to the degra-dation of underwater image quality,this paper combines the underwater image enhancement method and the color correction method to improve the quality of underwater images.In this paper,our works can be summarized as follows.Firstly,aiming at the problem of insufficient number and diversity of underwater image samples,a network combining depth image reconstruction and underwater image generation is proposed.The underwater image is simulated by style migration.Compared with other image simulation methods,the proposed method fully considers the attenuation and scattering effects of light propagation in water,and can obtain pictures closer to underwater style with shorter training time.Then,aiming at the problem of underwater image color distortion,this paper proposes a dynamic threshold color correction method based on image global information combined with the loss law of light propagation in water.Compared with other color correction methods,pro?posed method can be applied to many underwater scenes such as low illumination?blue-green and other.Finally,for the purpose of improving the overall sharpness of the underwater images.the color-corrected image is reconstructed by multi-scale recursive super-resolution convolutional neural network to achieve the overall enhancement.The results on several public state-of-the-art methods show that the proposed methods can simultaneously recover the color information and texture details of underwater images in various scenes,thus achieving color correction and overall clarity of underwater images.