摘要： 侧扫声纳图像边缘检测较困难，为此，提出一种针对该图像特点的多尺度边缘检测方法。对侧扫声纳图像进行非下采样Contourlet变换(NSCT)分解，根据斑点噪声在NSCT域的分布特点，进行局部自适应去噪。通过各方向子带沿边缘方向的插值和非极大值抑制寻找模极大值点。通过类内方差最小化法自适应确定阈值，由阈值处理得到各子带的边缘。经边缘融合实现完整的边缘图。实验结果表明，该方法具有边缘检测完整、定位准确、伪边缘点少等优点。

关键词: 侧扫声纳图像, 非下采样Contourlet变换域, 去噪, 模极大值, 边缘融合, 边缘检测

Abstract: To solve the problem of the difficulty in side-scan sonar image edge detection, a muti-scale edge detect method based on the characteristic of side-scan sonar image is proposed. Side-scan sonar image is decomposed in Non Sampling Contourlet Transform(NSCT) domain and image is denoised locally and adaptively according to the characteristic of speckle noise in NSCT domain. Maximum modulus points are found by interpolation in the direction of edge and non-maximum suppression. The threshold is automatically determined based on minimum interclass variance algorithm and the edge of each subband is acquired by thresholding. The binary edge map is obtained by edge fusing. Edge detection results show that the proposed method has the advantages of edge integrity, positioning accuracy and fewer false edge points.

Key words: side-scan sonar image, Non Sampling Contourlet Transform(NSCT) domain, denoising, modulus maximum, edge fusion, edge detection

中图分类号: 

• TP391.41