自适应可变结构元素形态学理论方法及其在高分影像分割中的应用研究

Morphology is an efficient tool for image processing, however, the nature image usually contains various objects with different shapes and sizes in most cases, when morphological operators are directly applied to such image with fixed shape and size structuring element, there maybe result in the objects position offset and weak target lost. For the purpose of processing image without objects contour position bias and preserving complete objects, the theory and method of adaptive variable structuring element morphology will be proposed, where the shape and size of the structuring elements are no longer fixed, each pixel corresponding to structuring element with different shape and size according to the current pixel region features. That is to say, the shape and size of the structuring element is adaptively changed with the object region structural characteristics. Once the adaptive variable structuring element set is quantitatively established and defined, the corresponding adaptive morphological basic and derivative operations can be defined for image processing and avoiding the object loss meanwhile maintaining the accurate location of the object edge. The implementation of the project desires to provide the theoretical methods for refined morphological image processing, improve and rich the morphological theory itself.. On the basis of the proposed theories and methods of the adaptive variable structuring element morphology, corresponding adaptive morphological smoothing simplification algorithms are explored and proposed to eliminate the over-segmentation of watershed segmentation, avoid the loss of weak objects, while maintain the advantage of watershed segmentation to produce closed contour and accurate positioning. It also desires to propose the full accurate segmental algorithm for high-resolution image segmentation by employing the combination of the proposed adaptive variable structuring element morphological smoothing simplification algorithm and the classical watershed transform, develop high-resolution image analysis software to achieve accurate statistical of the various resources using high-resolution images, and provide technical support for the governments, departments and enterprise to make scientific policy decisions.

形态学是一种有效的图像处理工具，但由于其采用单一形状和大小的结构元素对图像进行处理，而多数情况下图像中目标形状和大小各异，致使处理后的目标轮廓容易发生位置偏移和弱目标丢失。为此项目拟提出自适应可变结构元素形态学理论方法，对不同像素依据其区域特征构造不同形状和大小的自适应可变结构元素，进而定义相应的自适应形态学基本运算和组合衍生运算对图像进行处理，在保持目标边缘轮廓准确定位的同时避免目标的丢失，为实现形态学图像处理的精细化和完善形态学理论自身提供方法。. 在理论方法的基础上，提出相应的自适应形态学平滑简化算法并结合分水岭变换对高分影像进行分割，通过平滑简化，消除造成分水岭过分割的干扰因素，保持目标间边缘轮廓的定位，避免弱目标的丢失，同时保持了分水岭分割能够产生封闭且目标轮廓定位准确的优势，实现高分影像中形状和大小各异目标的准确分割，进而为不同资源目标的准确分析统计提供技术保障.

利用形态学运算对遥感图像进行分割时，由于地物目标的多样性和复杂性，直接分割容易产生过分割，通常采用形态学滤波平滑后能够产生较为满意的分割效果，但存在目标边界轮廓发生偏移和目标丢失的问题，原因在于经典形态学运算采用单一形状和大小的结构元素对图像进行处理，致使处理后目标轮廓容易发生位置偏移和小目标丢失。为此项目提出了自适应可变结构元素形态学理论方法，构建了不同形状和大小的自适应可变结构元素，定义了相应的自适应形态学基本和组合衍生运算，在保持目标边缘轮廓准确定位的同时避免了目标的丢失，为实现形态学图像处理的精细化和完善形态学理论自身提供方法。在理论方法研究的基础上，提出相应的自适应形态学分割方法对高分影像进行分割，实现了高分影像中各种形状和大小各异目标的准确分割，为不同地物资源目标的准确分析统计提供了技术方法参考。. 项目的主要研究内容和成果如下：. （1）通过图像特征属性构建了形状和大小可自适应变化的椭圆结构元素，提出了基于椭圆结构元素的自适应形态学运算模型；（2）构建了基于邻域灰度差矢量场和相对密度自适应结构元素构造方法；(3）提出了基于最小生成树的显著性自适应形态学结构元素构造方法；（4）定义了基于不同类型自适应结构元素的形态学运算；(5) 通过自适应可变结构元素形态学运算实现图像平滑滤波运算，提出了基于GAN自适应形态学结构元素构造方法及图像增强方法；（6）提出了基于不同类型自适应结构元素的形态学图像处理方法；（7）提出了基于自适应形态学运算结合其他理论的遥感图像目标分割方法。同时项目也利用其他理论对遥感图像分割分类方法进行了探索和研究。. 自适应形态学运算由于采用了大小和形状随图像目标特征的结构元素，可以实现对图像目标的精细处理，能够保持图像处理后的目标轮廓和定位保持不变，可以应用于图像增强、图像分割等对图像目标定位要求较高的应用场合。但研究也发现自适应形态学运算的复杂性较高，实时性有待进一步提高。