一种多方向的合成孔径雷达图像边缘检测方法

A Multidirectional Edge Detection Method for SAR Image

摘要:

经典的边缘检测方法(Canny, Roberts, Prewitt等)受合成孔径雷达(SAR)图像斑点噪声的影响致使其检测效果不佳；基于SAR图像的假设检验方法(Ratio算法等)其检测性能不随图像强度的变化而变化,边缘检测错误率低,但检测到的边缘片段较粗;针对上述问题,结合小波的多分辨分析特性以及Contourlet变换的充分捕获高维奇异信号的能力,提出一种改进的Contourlet变换与改进小波变换的模极大值方法对SAR图像进行边缘检测,在16个方向上对SAR图像进行边缘提取,进而保证了边缘信息的丰富性;试验结果表明,该方法提取的SAR图像边缘位置精确,运算速度较快,利于后续的图像匹配。

Abstract:

Because of the influence of SAR image’s coherent noise, the classic method of edge detection (Canny, Roberts, Prewitt, etc.) does not apply to the SAR image. Detection performance of the method of hypothesis testing based on SAR image (Ratio algorithm, etc.) does not change following the changing of image intensity, and edge detection error rate is low, but the detected edge segment is thick. To solving these questions, a modulus maxima method combined with improved Contourlet transform and improved wavelet transform is presented to detect SAR image edge in 16 directions, using the characteristic of wavelet multiresolution analysis and the capability of Contourlet transform to fully capture high dimension singular information of image. The test results show that the location of exacted edge of the SAR image is precise, and the speed of computing is fast, which is help to the subsequent image matching.