融合多尺度统计信息模糊C均值聚类与Markov随机场的小波域声纳图像分割

发布时间：2018-02-05 23:26

  本文关键词： 信息处理技术 声纳图像分割 模糊C均值聚类 Markov随机场 小波域 迭代条件模型算法　出处：《兵工学报》2017年05期 　论文类型：期刊论文

【摘要】：声纳图像成像质量差、特征信息弱,目标分割存在一定困难,为此提出一种融合多尺度统计信息的模糊C均值(FCM)聚类与Markov随机场(MRF)的小波域声纳图像分割算法。小波域中低频信息统计特性描述了低频不同区域像素聚类情况,高频信息反映了该方向纹理特征,依据低频子带的统计峰值选取FCM初始聚类中心,应用小波域FCM聚类算法对声纳图像进行预分割,抑制噪声的影响,提高了预分割的准确性;构建初分割后图像的多尺度MRF模型,尺度间节点标记的相关性采用1阶Markov性表征,尺度内构建2阶邻域系统描述系数间的标记联系,标记场采用双点多级逻辑模型建模,同一标记的系数特征场采用高斯模型建模,弥补了MRF算法中层次信息和轮廓信息描述的不足;应用迭代条件模型算法求其最小能量下的标记场,实现声纳图像分割。从视觉主观效果和客观评价指标两方面的实验结果验证表明,该算法分割声纳图像均优于FCM聚类算法和MRF算法,分割的声纳图像边缘与细节的清晰度、精细度均有一定程度改善。
[Abstract]:Sonar image imaging quality is poor, feature information is weak, target segmentation is difficult. In this paper, a fuzzy C-means (FCM) clustering and Markov random field (MRF) clustering for multiscale statistical information are proposed. Wavelet domain sonar image segmentation algorithm. The statistical characteristics of low frequency information in wavelet domain describe the low frequency pixel clustering in different regions. The high frequency information reflects the texture feature of this direction. According to the statistical peak value of the low frequency sub-band, the initial clustering center of FCM is selected, and the wavelet domain FCM clustering algorithm is used to pre-segment the sonar image to suppress the influence of noise. The accuracy of presegmentation is improved. The multi-scale MRF model of the first segmentation image is constructed. The correlation of the node markers between scales is characterized by the first-order Markov, and the marker relation between the describing coefficients of the second-order neighborhood system is constructed within the scale. The label field is modeled by two-point and multi-level logic model, and Gao Si model is used to model the coefficient characteristic field of the same marker, which makes up for the deficiency of the description of hierarchical information and contour information in MRF algorithm. The iterative conditional model algorithm is used to find the label field under the minimum energy, and the sonar image segmentation is realized. The experimental results from the visual subjective effect and the objective evaluation index show that the proposed method can be used to segment the sonar image. This algorithm is better than FCM clustering algorithm and MRF algorithm, and the edge and detail of the sonar image segmentation are improved to some extent.
【作者单位】： 三峡大学水电工程智能视觉监测湖北省重点实验室;三峡大学计算机与信息学院;
【基金】：国家自然科学基金重点项目(U1401252);国家自然科学基金项目(61272237) 水电工程智能视觉监测湖北省重点实验室开放基金项目(2015KLA05)
【分类号】：TB566;TP391.41
，

本文编号：1493031