Key Technology Research On Image Restoration For Imaging Through Water-air Interface From Underwater Based On Random Distortion Compensation

Pelicans and terns foraging above water,and dive into the water to hunt their preys;archerfishes shoot water jet accurately from underwater to hunt flies above water;jellyfish has upper lens eyes for observing the panorama of the sky through Snell window,helping them to pass through mangrove swamps.These creatures live on their special vision through water-air-interface.A similar research area on imaging through water-air interface is growing due to the demand of military applications.Virtual periscope is a novel passive underwater imaging system,which observes the above-water scenes for ships and airplanes without exposing its saliency.Based on the application requirement of virtual periscope,this thesis delve into the through-interface imaging key technology from the perspective of image processing.The main research works accomplished in the thesis are as follows:(1)Optical phenomena concerning imaging through water-to-air Snell window are analyzed using numerical simulation.A physical model of imaging through water-toair Snell window under calm water condition is studied and used to transform a circular Snell window to a rectangle azimuth-zenith-coordinated panorama.(2)The image degraded by turbulent water-air-interface is analyzed qualitatively,and classified into 3 patterns: 1.Blur due to motion and anisoplanatism;2.Random reflection distortion;3.Internal total reflection dark area enters the Snell window tearing image into pieces,which is usually seen under wind-generated capillary wave.Based on a special histogram feature exposed in those “destroyed” images,a frame decimation algorithm is used to filter out destroyed images.For the remaining frames,a degradation model concerning blur,distortion and noise is proposed.(3)A warping-field-based distortion model is proposed,and the numerical methods using warping field are proposed and formulized.(4)Image distortion caused by turbulent water-air interface is much stronger than that of which caused by atmospheric turbulence.The state-of-art turbulence compensation methods used to tackle atmospheric turbulence is not good enough to tackle the large displacement caused by water.Several improved algorithms are proposed in the thesis.And they are demonstrated,compared,and analyzed through both synthetic data and real through water images.(5)Distortion compensation methods rely on a large quantity of frames,and usually take multiple stages to achieve robust restoration results,which makes them unable to perform real-time processing.Thus,a progressive restoration method,synthetic imaging with centroid evolution,is proposed.This method can work synchronously with image acquisition,mitigate both blur and distortion progressively.(6)An underwater imaging experiment platform is built to capture through-water scenes from underwater.The turbulent image data is acquired and used to demonstrate the proposed progressive turbulence-mitigation method in video.