模糊聚类的侧扫声纳图像分割算法

Segmentation algorithm of fuzzy clustering on side scan sonar image

针对侧扫声纳图像的特点,利用二维经验模态分解(BEMD)将图像分解成若干固有模态函数(IMF)和1个余量.分析了侧扫声纳图像背景区的图像频率特征,通过增强目标和阴影的特征信息降低噪声的影响.提取图像的高斯马尔可夫纹理(GMRF),用来表达图像像素点间的空间关系,以减少图像的误分割.利用BEMD和GMRF改进模糊C均值聚类算法,提出了新的聚类准则和距离函数,形成一种新的模糊聚类算法.利用该算法对不同的侧扫声纳图像进行分割,并将分割结果与其他典型的聚类算法的分割结果进行比较,验证了该算法的抗噪性和准确性.

Aiming at the characteristics of background areas in the sonar imagery,the side scan sonar image was decomposed into several intrinsic mode functions（IMFs） and a residue with the bidimensional empirical mode decomposition（BEMD）.Features of the targets and shadows were enhanced and the noise was reduced.In order to use the spatial relationships between pixels,Gauss Markov random field（GMRF） was used to extract the local texture,and the mis-segmentation was reduced.Finally,a new fuzzy clustering algorithm was proposed,and its clustering criterion and distance function were rewritten.By the comparison with other traditional clustering methods in sonar image segmentation field,the proposed algorithm was proved to be superior to others.