基于NSST和改进数学形态学的遥感图像目标边缘提取

Target Edge Extraction of Remote Sensing Images Based on Non-Subsampled Shearlet Transform and Improved Mathematical Morphology

为了从遥感图像中提取出更为准确完整的目标边缘,提出一种基于无下采样Shearlet模极大值和改进数学形态学的目标边缘提取方法。首先采用无下采样Shearlet变换(NSST)将图像分解成边缘细节丰富的高频分量和边缘细节较少的低频分量;然后结合不同分解程度下边缘像素点处的系数关系,对高频分量的各个子带进行模极大值检测,再经过双层掩膜筛选后得到高频边缘提取结果;对低频分量采用改进的数学形态学方法,得到低频边缘提取结果;最后将上述两部分融合,使用区域连通方法去除孤立点,得到最终的目标边缘图像。大量实验结果表明,与Canny以及其他4种同类边缘提取方法相比,本文方法所得边缘定位准确、完整清晰、细节丰富,且抗噪能力强,为后续遥感图像目标特征提取与识别奠定更好基础。

In order to extract edges of target area more completely and accurately from remote sensing images,a method of target edge extraction is proposed based on improved mathematical morphology and modulus maxima of non-subsampled Shearlet transform.Firstly,the image is decomposed into high-frequency components with more edges and details and low-frequency component with fewer edges and minutiae through non-subsampled Shearlet transform.Then considering the property of coefficients of edge points under different decomposing conditions,the modulus maximum detection is performed for each sub-band of high-frequency components and the double-layer mask is adopted afterwards so as to get the high-frequency edge extraction result.Moreover,the low-frequency component is processed through the improved mathematical morphology method to get the low-frequency edge extraction result.Finally,the above two parts are fused and the final target edge image is obtained after removing the isolated points according to the regional connectivity.A large number of experimental results show that,compared with Canny method and four similar edge extraction methods,the detected edges by the proposed method are accurate,clear,complete and with abundant details.The method has strong anti-noise performance,which lays a better foundation for the following target feature extraction and recognition of remote sensing images.