Research Of Underwater Acoustic Image Segmentation Based On Spectral Clustering Technique

Spectral clustering segmentation technique is a cluster analysis method to seek theoptimal partitioning of the data set based on the spectral graph theory. In recent years imagesegmentation based on spectral clustering has become a research focus in the field of imageprocessing. The main content of this paper is to achieve the spectral clustering segmentationfor the underwater acoustic image.Firstly, set up a segmentation model suit for underwater acoustic image and find theoptimal pretreatment method according to the criteria of maintaining the smooth edges of theregion with the characteristics of the underwater acoustic image; built the framework of thespectral clustering segmentation algorithm based on the image with the core of theconstruction of the similarity matrix.Secondly, on the basis of the basic model, propose a multi-level spectral clusteringsegmentation method based on area with the thought of optimizing spectral clusteringsegmentation algorithm performance and reduce the computational complexity, transform theobject of clustering from pixel into area. Get the sub-area with consistency by using the twoways of watershed and mean shift and subdivide the area with the thought of spectralclustering. On this basis, discuss the request of the parameters for the two types of spectralclustering based on pixel and based on area, then give a rational range of options. Combinethe thought of fuzzy clustering and spectral clustering segmentation, and promote the idea ofclustering with the thought of fuzzy mathematics to achieve the soft division of the data set.Finally, in order to obtain the quantitative analysis of the experimental data, discuss theevaluation means and methods of analysis the performance of the image segmentationalgorithm, design the objective and quantitative evaluation parameters.