Method Study On Image Segmentation Of Water Surface Target

Image guided missile is widely used in modern ocean battles for its strong independence and high guidance precision.In image guided technology,accurate segmentation of the target is the key to attack the target accurately and efficiently.It is the basis and premise to recognize the targets correctly.So,result of the segmentation directly determines the precision of the image guided missile.How to segment the target accurately and efficiently has always been a focus for scholars at home and abroad.Study of the technology of segmenting the target under complex conditions is both challenging and practical.This thesis carries out a study about image segmentation of the water surface target.Difficulties of the study mainly lie in: low contrast between the target and background,target is small whereas the background occupies a large area,fuzzy target edge and the existence of noise and so forth.In view of the problems mentioned above,main work and study of the thesis are as follows:Firstly,this thesis pre-processes the image.To solve the problem of image degradation and noise,this thesis studies several methods of image restoration in common use.Result of the experiment demonstrates that Wiener filtering is the best choice for image restoration.To solve the problem of image noise,analysis and comparison among median filtering,neighbourhood average filtering and histogram equalization are made.Result of the experiment shows that median filtering is relatively a better choice.Secondly,this thesis discusses image segmentation methods of different type,they are the foundation and theoretical basis for the improvement of segmentation algorithm.For the threshold segmentation holds the advantages of simple calculation,low complexity of the algorithm and good real-time.Take the characteristics of the image into consideration: this thesis takes the maximum entropy threshold segmentation to segment the image.For the one-dimensional maximum entropy only considers the gray level of the image and ignores the spatial information,the result of the segmentation is not so good.Thirdly,the thesis discusses two-dimensional maximum entropy which considers both the gray level of the image and the spatial information,the result of the segmentation is relatively improved compared with the one-dimensional maximum entropy,but doesn't meet the needs of good real-time,segmentation of the detailed information is insufficient,and the ability to suppress the noise is poor,optimization of it was discussed in this thesis.For the genetic algorithm has the merit of robustness,parallelism,adaptability and high convergence rate,and is widely used in many scientific research fields,this thesis combines the genetic algorithm and two-dimensional maximum entropy to optimize the process of searching the threshold.Experimental results show the optimized algorithm is better in real-time,algorithm complexity is relatively low,and segmentation of the target is pretty good.It is easy to apply this algorithm into practical use in engineering.And finally,from the perspective of engineering application,this thesis develops a software platform based on MATLAB GUI to segment the image which has the function of segmenting the image quickly with the improved segmentation algorithm proposed in the above chapter.