摘要： 针对侧扫声纳图像斑点噪声强的特点，提出一种改进的Canny算子进行边缘检测. 根据斑点噪声的乘性模型和瑞利分布特性，在非下采样Contourlet变换域进行局部自适应降斑. 该方法在有效抑制斑点噪声的同时可较好地保护边缘，避免了Canny算子造成的边缘模糊. 计算降斑后图像的梯度值分布，对梯度幅值进行非极大值抑制得到极大值点. 将梯度模的极大值点分成强边缘点、弱边缘点与非边缘点3 类，基于类间方差最大自适应确定区分3 类的双阈值，经双阈值处理与弱边缘连接得到边缘图. 对模拟声纳图像和实际声纳图像的边缘检测结果表明，较之Canny算子和小波模极大等边缘检测方法，该方法具有边缘检测完整、定位准确、伪边缘点较少等优点.

To deal with strong speckle noise of side-scan sonar images, an edge detection method based on improved Canny operator is proposed. According to a multiplicative model and the Rayleigh distribution of speckle noise, sonar image de-speckling is performed locally and adaptively in the contourlet transform domain without sub-sampling. This effectively suppresses speckles and better protects edges without blurring due to Gaussian smoothing. Gradients of the de-speckled sonar image are computed. The maximal magnitudes of the gradients are obtained by non-maximum suppression. The maximal points are classified into three types: strong edge points, weak edge points and non-edge points. Two thresholds are automatically determined based on the maxima of inter-class variance. A binary edge map is obtained with the two thresholds followed by weak edge linking. Experiments on both synthetic and real sonar images show that the proposed method has the advantages over other methods such as Canny operator and wavelet modulus maxima in terms of edge integrity, positioning accuracy and false edge points.