Underwater Optical Image Enhancement And Data Generation Algorithms

As an important component of natural resources,Marine resources play an irreplaceable role in the entire resource system and are showing increasingly important economic,military,and social values.The prerequisite for fully and reasonably utilizing marine resources is to have advanced underwater exploration technology.At present,underwater optical imaging is one of the main means to obtain underwater information,and this method plays an important role in close range underwater detection.However,compared to other images,underwater optical images face serious problems such as image quality degradation and data scarcity.Therefore,research on underwater image quality restoration and data augmentation is an urgent problem to be solved at present.This article focuses on two aspects of underwater image visual enhancement and underwater style synthesis.Firstly,due to the severe absorption and scattering of visible light by water bodies,the underwater images directly obtained by the camera exhibit characteristics of color shift,fog blur,and detail loss.This article proposes a color correction strategy based on Lab color space by considering the perceptual dependence between colors and the absorption characteristics of underwater light.This method does not require data-driven and can restore the natural color appearance of images in the vast majority of underwater scenes.On this basis,this article also proposes a simple and effective depth estimation method for a single underwater image by utilizing the attenuation characteristics of underwater images.Based on the obtained depth information,the details of underwater image loss can be further restored.Secondly,in the context of the rise of deep learning,due to high acquisition costs and diverse attenuation styles,the existing underwater image data is limited in quantity and style,which limits the development of many downstream visual tasks based on deep learning methods.This article proposes an underwater image synthesis method based on differential attenuation synthesis to expand underwater image data.Considering the true attenuation characteristics and diverse attenuation styles of underwater images,this article adds neural style transfer technology to the physical imaging model.This method can synthesize underwater effects from multiple non underwater images based on a real underwater image.This article uses a single network to achieve diversified attenuation effect synthesis,and the synthesized results have nearly real underwater light attenuation characteristics.This article uses rich evaluation indicators and conducts experiments on a large number of comparative algorithms.The experimental results show that the two methods proposed in this article have better performance in their respective tasks.