Study On Underwater Image Segmentation Technique

As an increasing urgency of human beings towards marine resources,the requirements for underwater environment detection technology were increasing.Underwater images were critical tools for recognizing and analyzing underwater environment.The processing technologies of underwater images used in underwater information processing were also critical.Compared with those ordinary images,underwater images often had uneven illumination,severe noise,and fuzzy details,which attributed to the light source and the environment during imaging.These features leaded to the difficulty in processing and recognizing underwater image.And the segmenting accuracy of underwater images was not high,which affects the subsequent research and application.As one of the basic computer vision technologies,image segmenting was an important step before high-level visual technologies such as image recognizing.It aimed at extracting the targets in images and preparing for the subsequent processing.Image segmenting had critical influence on the results of the whole image processing.The smoothing and enhancement algorithms of underwater images were researched and used to pre-process underwater images.The results indicated that every pre-processing method had special advantage.The edge-preserving mean filtering method and the edge-preserving median filtering method had advantage on removing impulse noise,while the bilateral filtering method could be used to remove Gaussian noise and the holomorphic filtering method could eliminate the foggy features of underwater image.Besides,the background illumination filtering had better performance on handling with illumination unbalance in images.Therefore,selecting an appropriate pre-processing method of underwater images was critical,and could benefit to image segmenting that is the next step of image processing.We performed researches on the underwater image segmenting algorithms based on the regional theory.The transition regions extracting algorithm based on the regional complexity theory,the multi-threshold algorithm,and the fuzzy C-Means algorithm were analyzed and used to the segmenting of underwater images.Then the above algorithms were improved by combining with the neutrosophy,and three new algorithms were proposed.These new algorithms were used to segment underwater images and some indexes were proposed to evaluate the segmenting results.These indexes included the area ratio,the gray-level energy of the non-reference image,the discrete entropy,the relative entropy,the mutual information,and the redundancy.The results of underwater images segmenting experiments and the calculated indexes indicated that these improved algorithms had better robustness to background noise,and their segmenting results were more concise and accurate.We also performed researches on the underwater image segmenting algorithms based on the edge theory.The contrastive research method was performed on the research of the segmenting algorithms based on edge,graph,and watershed theory.At first,the principles of the above image segmenting algorithms were analyzed.Then these algorithms were applied to the segmenting of underwater images.The research focused on how to improve performance of the watershed algorithms by combining with the neutrosophy.The improved watershed algorithm was used to the segmenting experiments of underwater images.The results suggested that the improved watershed algorithm that combined with the neutrosophy could avoid excessive segmenting to underwater images and keep the edge of target region complete and positioning accurate.