Underwater Image Restoration Method Based On Image Registration

Since the 21 st century,the sustainable development of human society has become more and more dependent on the ocean.It has become a major issue of the development of the country to accelerate the development and utilization of marine resources.Imaging is an intuitive and effective way of expressing underwater information.However,because the underwater environment is complicate and the movement of water is random,the underwater images often suffer from serious distortions.Especially at the air-water interface,imaging through a fluctuating air-water surface is a challenging task since the light rays bent by unknown amounts lead to complex geometric distortions.On account of this,not only the overall visual effect of images will be reduced,but the subsequent tasks,such as pattern recognition and target detection,will be hindered.During the research on the restoration of trans-media underwater distorted images,this thesis has mainly completed the following tasks:(1)The restoration methods of trans-media underwater distorted images at home and abroad are organized and summarized.(2)The optical imaging principle and imaging characteristics of trans-media underwater distorted images are studied.(3)Two classical trans-media underwater image restoration methods are studied.Through simulation analysis,it is found that using a blurry average image as the reference image will severely limit the effect of image registration.(4)Based on the iterative image registration method combining the B-spline transformation model and the free deformation model,it is proposed to improve the effect of image registration by reconstructing high-quality reference images.Reconstructing the reference image is mainly composed of two steps.The first step is the lucky patches fusion technology,which uses the structural similarity index as the evaluation method of geometric distortion.After the image patches with less distortion are selected from the sequence,the patches are fused and stitched into a single less distorted image.The second step is the blind deconvolution technique,which combines the Richardson-Lucy algorithm and the iterative blind deconvolution algorithm.This step is applied to remove the image blurs caused by the lucky patches fusion.(5)The algorithm of this thesis is tested using the real trans-media underwater warped video data,and compared with the aforementioned classic methods.The experimental results show that integrating the lucky patches fusion and blind deconvolution into the iterative image registration framework can not only obtain better restoration effect,but consume less time and have higher operation efficiency.