基于数学形态学的遥感图像分割算法研究

发布时间：2018-04-28 13:11

本文选题：遥感图像 + 数学形态学　；参考：《成都理工大学》2014年硕士论文

【摘要】：在图像处理过程中,人们往往对图像中某些部分感兴趣,而不同的目标在图像中一般具有不同的特征。为了识别和分析图像中的目标区域,就需要把它们从图像中分离开来。遥感图像通常分辨率较高,包含信息丰富,目标结构比较复杂,同时也含有大量的噪声。由于遥感图像的这些特性,使得传统数学形态学的图像分割算法并不能完全适用,因此在一定程度上阻碍了图像分割技术在遥感领域的推广和应用。本文针对遥感图像的特性,结合数学形态学在图像分割中的应用,提出了一种基于数学形态学的改进分水岭分割算法。通过实验验证,该算法具有高效、准确、快速等特点,可以得到连续封闭的目标区域。由于传统分水岭分割算法对噪声比较敏感,将该算法直接作用在遥感图像中会产生过分割现象,分割的效果并不理想。本文的主要工作包含以下几个方面：首先,本文以二值形态学为基础,介绍了数学形态学的膨胀和腐蚀基本运算,并扩展到灰度形态学当中。通过几种梯度算子的分析和比较,本文提出的多尺度形态学梯度算子能够得到较为理想的梯度图像。其次,采用扩展最小变换对梯度图像进行标记时,阈值的选取通常是人工设定的,一般带有一定的盲目性。本文采用二维Otsu阈值分割算法自适应地获取最佳阈值,避免了人为的干预。实验表明,该算法可以有效地抑制噪声的干扰,能够标记出图像的主要轮廓,保持完整的信息。最后,本文提出标记分水岭分割算法,克服传统分水岭算法存在的过分割缺陷。对输入的原始图像进行形态学滤波处理,在得到的梯度图像上提取标记,并把标记强制作为极小值修改梯度图像。从实验的结果可以看到,该方法有效地解决了分水岭算法的过分割问题,得到了较好的分割效果。
[Abstract]:In the process of image processing, people are often interested in some parts of the image, and the different targets usually have different features in the image. In order to identify and analyze the target area in the image, it is necessary to separate them from the image. The remote sensing image usually has a high resolution, contains rich information, and the structure of the target is more complex. It also contains a lot of noise. Because of the characteristics of remote sensing images, the traditional mathematical morphology image segmentation algorithm can not be fully applied, so to a certain extent, it hinders the popularization and application of image segmentation technology in the field of remote sensing.
In view of the characteristics of remote sensing images and the application of mathematical morphology in image segmentation, an improved watershed segmentation algorithm based on mathematical morphology is proposed. Through experiments, it is proved that the algorithm has the characteristics of high efficiency, accuracy and fast speed and so on. It can get the continuous closed target area. Because the traditional watershed segmentation algorithm has the noise ratio to the noise ratio. It is more sensitive to use the algorithm directly in remote sensing image to produce over segmentation, and the effect of segmentation is not ideal. The main work of this paper includes the following aspects:
First, based on the two value morphology, this paper introduces the basic operation of the expansion and corrosion of mathematical morphology and extends to the gray scale morphology. Through the analysis and comparison of several gradient operators, the multi-scale morphological gradient operator proposed in this paper can get a more ideal gradient image.
Secondly, when using the extended minimum transform to mark the gradient image, the selection of threshold is usually artificial, and it usually has a certain blindness. In this paper, a two-dimensional Otsu threshold segmentation algorithm is used to obtain the best threshold and avoid human interference. The experiment shows that the algorithm can effectively suppress the noise interference and can be labeled. Remember the main outline of the image and keep the complete information.
Finally, this paper proposes a marked watershed segmentation algorithm to overcome the oversegmentation defects in the traditional watershed algorithm. The original image is processed by morphological filtering, and the labeling is extracted on the obtained gradient image, and the mark is forced to modify the gradient image as a minimum. It can be seen from the experimental results that the method is effectively solved. The over segmentation problem of watershed algorithm is proved to be effective.

【学位授予单位】：成都理工大学
【学位级别】：硕士
【学位授予年份】：2014
【分类号】：TP751

【参考文献】