基于梯度算子和类间方差边缘检测算法研究

发布时间：2018-05-12 13:15

  本文选题：边缘检测 + 二维最大类间方差算法　； 参考：《青岛理工大学》2016年硕士论文

【摘要】：图像边缘检测是图像分析和理解的基础,是图像特征提取与目标识别的重要前提,其结果的完整性直接影响到计算机对目标图像的理解和认知,因此图像边缘检测在计算机视觉及图像处理领域中具有至关重要的作用。通过研究一阶、二阶的微分算子和Canny算子,重点分析了它们对悬浮微粒的边缘提取效果,依据悬浮微粒的水下成像特点,提出以下三个方面的创新。(1)针对在复杂背景下Sobel算子提取悬浮微粒边缘时存在边缘粗糙和阈值精确度低的问题,提出了一种基于阈值分割与梯度算子相结合的图像边缘检测算法,该算法采用牛顿迭代算法实现快速收敛至最佳阈值。实验表明,该算法对悬浮微粒边缘的检测精度较高,且边缘线清晰准确。(2)由于水下照明导致悬浮微粒的成像过程中存在水平条纹噪声,针对传统梯度算子敏感性高、易检测出噪声的问题,提出了一种基于图像边缘增强的二维最大类间方差(2D-Otsu)边缘检测算法,该算法采用梯度锐化算子消除水平条纹噪声。实验结果表明,该算法可在一定程度上滤除水平条纹并检测出悬浮微粒的边缘特征,对水平条纹噪声具有较强的鲁棒性。(3)针对传统Canny算子需人工设定阈值范围、自适应能力较弱的问题,提出了一种基于二维Otsu自适应阈值的Canny边缘检测算法。实验结果表明,该算法解决了噪声滤除问题和边缘有效信息保留之间的矛盾,具有较强的鲁棒性和适应性。本文提出的边缘检测算法解决了传统的梯度算子提取悬浮微粒边缘时的部分问题,比传统算法具有检测精度高、定位准确度高、适应性和鲁棒性强的优点。
[Abstract]:Image edge detection is the basis of image analysis and understanding and an important premise of image feature extraction and target recognition. The integrity of the result directly affects the understanding and cognition of the target image by computer. Therefore, image edge detection plays an important role in computer vision and image processing. By studying the first and second order differential operators and Canny operators, the effect of edge extraction on suspension particles is analyzed, according to the underwater imaging characteristics of suspended particles. In view of the problem of rough edges and low threshold accuracy, the Sobel operator can extract the edge of suspended particles in complex background. An image edge detection algorithm based on the combination of threshold segmentation and gradient operator is proposed. Newton iterative algorithm is used to rapidly converge to the optimal threshold. Experimental results show that the proposed algorithm has a high accuracy for the edge detection of suspended particles, and the edge lines are clear and accurate. (2) because of the horizontal fringe noise in the imaging process of suspended particles caused by underwater illumination, the traditional gradient operator is highly sensitive. The problem of noise detection is easy to be detected. A 2D maximum inter-class variance 2D-Otsu-based edge detection algorithm based on image edge enhancement is proposed. Gradient sharpening operator is used to eliminate horizontal fringe noise. The experimental results show that the algorithm can filter out horizontal stripes to some extent and detect the edge features of suspended particles. The algorithm is robust to horizontal fringe noise. For the problem of weak adaptive ability, a Canny edge detection algorithm based on 2-D Otsu adaptive threshold is proposed. The experimental results show that the algorithm solves the contradiction between the noise filtering problem and the edge effective information retention, and has strong robustness and adaptability. The edge detection algorithm proposed in this paper solves part of the problem when the traditional gradient operator is used to extract the edge of suspended particles, which has the advantages of high detection accuracy, high localization accuracy, adaptability and robustness.
【学位授予单位】：青岛理工大学
【学位级别】：硕士
【学位授予年份】：2016
【分类号】：TP391.41
，

本文编号：1878755