【作者】 景雨；

【导师】 安居白；

【作者基本信息】 大连海事大学 ， 计算机应用技术， 2011， 博士

【摘要】 随着我国航海事业、海洋石油开发和沿海经济的迅速发展,溢油事故和不合法的废油排放频频发生。这些问题严重地威胁着海洋环境并造成巨大的经济损失。所以,溢油的及时监测、识别、回收和清理工作显得十分重要。遥感技术的迅猛发展使得人们通过遥感手段来监测和识别溢油成为可能,大量针对遥感数字图像处理的技术也应运而生。其中,图像边缘检测技术显得尤为重要,因为无论溢油的识别,位置的确定或者溢油量的获取,我们都需要首先确定溢油区域的边界信息。由于溢油遥感图像通常存在大量的条纹、斑点噪声,边界模糊,溢油区和海水区对比度较低以及灰度不均匀性等问题,使得现存的边缘检测方法很难获得精确的边缘检测结果。本课题以溢油遥感图像为重点研究对象,对图像边缘检测的发展情况,现状进行了深入、系统的探讨和研究,创新性地提出了3种边缘检测算法。具体工作可概括如下：1.借鉴了传统边缘检测方法的思路即：候选边缘点的确定,阈值去噪和边缘连接三个过程,且针对阈值去噪过程中常用的全局阈值算法的缺陷提出了一种动态分块阈值去噪算法,该算法考虑了局部边缘的梯度信息对阈值的影响,避免了在使用全局阈值算法时由微弱精细边缘形成的局部极大值会随着由灰度不均匀、噪声等产生的极大值一起被滤除掉,从而更加准确的确定了真实的边缘点,且对候选边缘图像中存在的伪边缘和噪声具有较好的抑制能力。在边缘连接的过程中,针对GDNI边缘连接方法在噪声和模糊边界的干扰下容易产生误连接的问题,提出一种改进的GDNI边缘连接算法,该算法综合利用了边缘终断点间的灰度信息,欧几罩德距离信息以及方向信息,实现了对终断边缘点的准确连接。实验结果表明,提出的两种算法的结合较好地实现了具有低对比度、弱噪声问题的溢油遥感图像的边缘检测,且具有很好的实时性。2.充分利用区域可扩展拟合模型RSF和全局最小化主动轮廓模型GMAC各自的优势,提出了一种新的基于区域可扩展的全局主动轮廓边缘检测模型(RSF-GAC)。该模型在全局最小化框架下引了图像的边缘信息和局部区域信息,既能够避免能量函数的演化陷入局部极小值,又能够抑制图像中存在的噪声、灰度不均匀性和低对比度等问题。在曲线演化和数字最小化的过程中,引入了基于加权全变分的对偶规则,将RSF-GAC模型的最小化问题转化成一个迭代过程,并对数字化迭代过程进行了文字描述。与传统的基于水平集的曲线演化方法相比,基于加权全变分对偶规则的最小化方法不需要定义初始轮廓,实现更简单,曲线收敛更快,更准确。实验结果证明,提出的RSF-GAC边缘检测模型对于具有低对比度、强噪声、弱灰度不均匀性问题的溢油遥感图像都能够获得较理想的边缘检测结果,且与其它算法相比,具有较少的控制参数、较高的精确性和较快的收敛速度。3.由于溢油遥感图像中会存在各种程度的灰度不均匀性的问题,而现存的主动轮廓边缘检测模型无法很好的处理该问题,基于此本文提出一种健壮的基于局部高斯拟合和灰度不均匀性纠正的主动轮廓边缘检测模型LGF-IHC。首先对具有灰度不均匀性的图像用数学方法来描述,并尝试建立一个基于局部高斯拟合和灰度不均匀性纠正的区域能量模型,然后结合测地线主动轮廓模型的边缘梯度信息构造了一个新颖、健壮的LGF-IHC主动轮廓边缘检测模型。从理论上借用了Chan的全局最小化优化思想来求得LGF-IHC模型的全局极小解。在数字最小化和曲线演化过程中,运用了基于加权全变分的对偶规则,快速、稳定地实现了LGF-IHC模型的全局最小化迭代过程。通过大量的实验证明了提出的LGF-IHC模型不仅能够有效的实现具有低对比度和强噪声问题的溢油遥感图像的边缘检测,还能够纠正图像中存在的各种程度的灰度不均匀性问题,从而得到理想的恢复图像。与基于动态分块阈值和改进的GDNI边缘连接的边缘检测算法以及基于RSF-GAC模型的边缘检测算法相比,基于LGF-IHC模型的边缘检测算法具有更高的精确性和更强的鲁棒性。算法的扩展证明了提出的LGF-IHC模型还能够处理目标和背景具有相同均值但不同方差的一类特殊图像,且在军事和医学图像中的应用进一步证明了LGF-IHC模型良好的可扩展性。更多还原

【Abstract】 With the current maritime cause, offshore oil development and rapid economic development along the coast, oil spills accidents and illegal oily discharge occur frequently on the sea, and these situations represent a serious threat to the marine environment and cause great losses of energy sources. Early detection, monitoring, containment, and cleanup of oil spill are crucial for the protection of the environment. It is possible to monitor and identify oil spill with the rapid development of remote sensing technology, and many image processing techniques of remote sensing have came into being. In particular, the technology of edge detection is an important tool for the location and acreage calculation of oil slick on the sea by aerial remote sensing. Whenever we need to identify oil spill, confirm the location or get the shape and acreage of oil spill, we have to get the edge information of oil slick images firstly. Due to the complexity of oil spill remote sensing image, it is very difficult to gain accurate edge detection results by conventional edge detection methods. Therefore, further research is still needed. In this research, we mainly focus on the edge detection of the oil spill remote sensing images, and put forward three innovative edge detection algorithms, the main works in this thesis can be summarized as follows:1. Considering conventional edge detection methods are focused on three main problems:determination of candidate edge points, threshold denoising and edge linking, we propose a new edge detection method based on a dynamic block threshold denoising algorithm and an improved GDNI edge linking algorithm. Comparing with the current common global threshold algorithm, the proposed dynamic block threshold algorithm considers the local information of edge gradient and avoids the problems while using global threshold, thereby determining the true edge points more accurately. In the process of edge linking, we propose an improved GDNI edge linking algorithm which uses a cost function based on the weighting combination of Euclidean distance, intensity information and angle information of edge ending points and finally to improve the edge linking decision. We demonstrate through several experiments that the combination of the two algorithms achieves good edge detection results for oil slick remote sensing images with low contrast and weak noise, and has a better real-time performance.2. A novel global active contour edge detection model (RSF-GAC) based on region scalable fitting is proposed, which makes full use of advantages of RSF model and GMAC model. RSF-GAC model introduces the edge information of image and local region information under the framework of GMAC, so RSF-GAC model can avoid the existence of local minima and meanwhile deals with the intensity inhomogeneity, noise, and weak edge boundaries exiting in images. In the process of the active contour evolving toward object boundaries and numerical minimization, a dual formulation based on the weighting total variation is used for converting the minimization problem of RSF-GAC into an iterative process and overcoming drawbacks of curve evolution method based on the usual level set and gradient descent method so that the process of minimization can be much easier and and our algorithm is independent of the initial position of the contour. The numberical iterative process has been described literally. Large numbers of experiment results have shown that the proposed RSF-GAC model outperforms other algorithms in terms of the efficiency and accuracy with satisfactory results of edge extraction for oil slick images with low contrast, strong noise and weak intensity inhomogeneity.3. Oil slick remote sensing images ususlly have a serious problem of intensity inhomogeneity. The existing active contour models can not deal with the problem very well, so a robust active contour model is proposed based on the local Gaussian fitting and the correction of intensity inhomogeneity. Firstly, we describe the image with intensity inhomogeinty by using a common mathematical model, and try to establish a region energy model based on the local Gaussian distribution. Finally a novel and robust LGF-IHC edge detection model is constructed by combining the established region energy model and the geodesic active contour model. In the process of seeking the minimum solution of LGF-IHC energy model, we use the Chan’s global minimum optimization theory. In the process of the active contour evolving toward object boundaries and numerical minimization, a dual formulation based on the weighting total variation is used and implements global minimization iteration of the LGF-IHC model fast and stably. Large numbers of experiment results have shown that the proposed LGF-IHC model can be robust to the high noise and severe intensity inhomogeneity existing in oil slick remote sensing images. In addition, the accurate edge detection of oil slick region and the correction of intensity inhomogeneity are simultaneously achieved via the proposed LGF-IHC model. Compared with the dynamic block threshold denoising algorithm and an improved GDNI edge linking algorithm and the RSF-GAC model, LGF-IHC edge detection model has higher accuracy and robustness, furthermore, it will have a better application prospect in the edge detection of oil slick remote sensing image.更多还原