【作者】 李长凯；

【导师】 杨学志；

【作者基本信息】 合肥工业大学 ， 信号与信息处理， 2010， 硕士

【摘要】 合成孔径雷达(SAR)具有全天时、全天候、高分辨率、成像技术灵活等特点,并且已经在军事和民用等领域获得广泛应用。SAR图像包含丰富的信息,通过数字图像分割技术可以从SAR图像中提取感兴趣区域或目标。图像分割是图像处理到图像分析的关键步骤,是实现特征提取、参数检测和模式识别等应用的基础,研究SAR图像分割技术具有很强的理论意义和实用价值。目前,基于区域化模型的SAR图像分割方法成为研究的热点,它比基于阈值、基于边缘检测等分割方法具有更强的抑制噪声能力,并且分割结果更精确。区域模型的建立存在过度分割和边缘检测两个问题,针对这两个问题,本文以瞬态系数(ICOV)梯度算子为基础,提出建立边缘保持的SAR图像区域化模型,并应用于SAR图像分割。为此,主要开展以下研究内容:1.研究建立具有边缘保持特性的SAR图像区域化模型。SAR图像的相干斑噪声会导致过度分割,并且破坏边缘信息。各向异性相干斑降噪(SRAD)滤波器不仅能有效滤除相干斑噪声以此减轻过度分割,而且具有保持边缘信息的能力,本文将SRAD应用到构建边缘保持特性的区域模型中。2.提出将ICOV梯度与分水岭变换相结合获得区域化模型的初始区域。ICOV梯度算子对边缘具有单一峰值响应,并且响应宽度很窄;而分水岭对弱边缘具有良好响应,能检测出连续、封闭且单像素宽度的边缘。将ICOV与分水岭相结合,在获得大量均匀区域同时,还实现了目标边缘的准确检测。3.研究构建基于分水岭初始区域的区域邻接图(RAG),完成SAR图像区域化模型的建立。传统的描述图像方式是基于像素水平的,数据量大而且图像结构复杂;而RAG以较小区域(即像素集)为基本单元描述图像上下文结构关系,简化了图像结构。最后,将建立的SAR图像区域模型与区域水平马尔科夫场(MRF)分割方法相结合,并应用于SAR图像分割中。实验结果表明,本文提出的分割方法比基于其它经典梯度的分割方法具有更高的目标边缘检测与定位性能。更多还原

【Abstract】 Synthetic aperture radar (SAR) can image in almost any condition and produce high-resolution images with good flexibility. SAR has been applied widely for military and civil use. People hope to extract the interesting regions by SAR image segmentation. Image segmentation plays a key role in digital image processing and works for characters extracting, paramerters measurement and pattern recognition. To study SAR image segmentation is meaningful to theories and applications.Compared with traditional segmentation based histogram and edge detection, the methods region-based is a popular tool due to its ability of suppressing noise and more accurate segmentation. But the methods region-based has to resolve over-segmentation and regions edge location. According to these, instantaneous coefficient of variation (ICOV) gradient operator is used to construct a region model characterized by edge-preserving, and the main work is as follows:1. To develop a SAR image region model characterized by edge-preserving. SAR image segmentation suffers from over segmentation and losing edge because of speckle. Speckle reducing anisotropic diffusion (SRAD) is an edge-preserving filter and used to reduce speckle and protect object edge in SAR images.2. To propose a method to get initial regions using instantaneous coefficient of variation (ICOV) gradient operator and watershed transform. ICOV has single peak response and narrow respose width at edge location. Watershed transform is able to detect continous and close edge with one-pixel width. The properties of ICOV and watershed transform are advantageous to edge detection.3. To construct region adjacency graph(RAG)based on initial regions of watershed transform. SAR image region model is accomplished by RAG. Traditional image is presentated based on pixels. RAG simplifies SAR image structures in the form of small regions (sets of pixels).At the end, the above region model of SAR image is combined with region-level MRF to be used in SAR image segmentation. The experiments results prove that the proposed segmentation leads to more accurate edge detection compared to the segmentation based on other classical gradient operators.更多还原