CAJ | 학술논문

针对目前高空间分辨率遥感影像分割预处理噪声去除过程中,通常都是对影像采用同一尺度,即同一尺寸的结构元素,进行滤波,忽略了不同地类中的噪声尺度不一致的问题.该文基于形态学开闭重建运算,采用加权思想,充分利用不同尺度结构元素能去除对应尺度噪声的特点,结合多个尺度结构元素的滤波结果,提出一种多尺度形态学滤波方法.试验结果表明,该方法能有效抑制由于滤波尺度选择不合适造成的影像“过分割”和“欠分割”问题,适合于对高空间分辨率遥感影像的多尺度噪声去除.
침대목전고공간분변솔요감영상분할예처리조성거제과정중,통상도시대영상채용동일척도,즉동일척촌적결구원소,진행려파,홀략료불동지류중적조성척도불일치적문제.해문기우형태학개폐중건운산,채용가권사상,충분이용불동척도결구원소능거제대응척도조성적특점,결합다개척도결구원소적려파결과,제출일충다척도형태학려파방법.시험결과표명,해방법능유효억제유우려파척도선택불합괄조성적영상“과분할”화“흠분할”문제,괄합우대고공간분변솔요감영상적다척도조성거제.
The morphological filters can suppress impulse noise or small image components/structures while preserving very important geometrical features such as edges. So, the morphological filters have been widely used in image preprocessing to remove the image noises and noise reduction is critical step for image segmentation. Morphological filters analyze the geometrical structure of image by locally comparing it with a predefined elementary shape called a structure element. Different scale image edges are detected by using several typical structure elements. Large amounts of experimental results demonstrate that the size of structure element have much dependence with image background. Therefore, many studies devote to the adaptive optimization of structure elements of morphological filters. However, the structure element of the same scale is traditionally adopted to establish a filter and remove noise from very high resolution satellite images prior to image segmentation. This method ignores the problem of inconsistencies between different land use types in the noise scale. In this paper, for the complicated background satellite imagery, a multi-scale morphological filtering method, which takes full advantage of the merits of large and small structure element by weighted strategy and combines them with the filtering results of multi-scale structure elements, is proposed based on morphological opening- and closing-reconstruction operations. To evaluate the multi-scale morphological filter for the image segmentation, three filtering approaches and segmentation accuracy assessment results are compared in this study. Qualitative and quantitative experimental results show that the proposed method can effectively solve over-segmentation and under-segmentation problem that result from improper scale of structure element. Compared with accuracy assessments of single scale and multi-scale morphological filters, the multi-scale morphological filter segmentation obtained higher accuracy than single scale filter segmentation, and is suitable for removing the multi-scale noise from very high resolution satellite images.