针对二维局部均值分解(bidimensional local mean decomposition,BLMD)中影响算法速度的两个主要因素:自适应搜索窗口和迭代终止条件,提出了优化方法,并在其基础上提出了一种边缘检测算法。该算法采用Delaunay三角剖分得到局部极值点的...针对二维局部均值分解(bidimensional local mean decomposition,BLMD)中影响算法速度的两个主要因素:自适应搜索窗口和迭代终止条件,提出了优化方法,并在其基础上提出了一种边缘检测算法。该算法采用Delaunay三角剖分得到局部极值点的理想规则化的三角网格,通过网格划分确定相邻极值点及滑动平均窗口的大小,并提出了一种新的BLMD算法迭代收敛条件,通过对人工合成图像以及自然图像的实验,证实了该优化算法与原算法结果非常接近甚至更优,且大幅度提高了计算速度。对BLMD得到的最高频分量进行直方图均衡,将其结果二值化,通过设定阈值剔除其中不连续的细小边缘,通过形态学将其骨骼化,得到最终提取的边缘。与几种典型边缘检测算子的比较实验表明,新算法可以较好地检测出图像边缘,相对于其他边缘检测算子,对于图像中的纹理等细节边缘有着更佳的检测效果;并且得益于BLMD图像多尺度分析的优势,较好地避免了因光照明暗等低频因素产生的假边缘,提取出的边缘更符合视觉上的主观检测。展开更多
A modified version of the Cotte, Lions, Morel and Coil theory for image selective smoothing and edge detection is proposed. Comparing with their model, the most important advantage of this modification is that the con...A modified version of the Cotte, Lions, Morel and Coil theory for image selective smoothing and edge detection is proposed. Comparing with their model, the most important advantage of this modification is that the convolution with Gaussian processes in the filtering process is replaced by solving an initial-boundary value problem for the heat equation, which simplifies the numerical scheme to some extent. Numerical experiments on natural images are presented for this model.展开更多
文摘针对二维局部均值分解(bidimensional local mean decomposition,BLMD)中影响算法速度的两个主要因素:自适应搜索窗口和迭代终止条件,提出了优化方法,并在其基础上提出了一种边缘检测算法。该算法采用Delaunay三角剖分得到局部极值点的理想规则化的三角网格,通过网格划分确定相邻极值点及滑动平均窗口的大小,并提出了一种新的BLMD算法迭代收敛条件,通过对人工合成图像以及自然图像的实验,证实了该优化算法与原算法结果非常接近甚至更优,且大幅度提高了计算速度。对BLMD得到的最高频分量进行直方图均衡,将其结果二值化,通过设定阈值剔除其中不连续的细小边缘,通过形态学将其骨骼化,得到最终提取的边缘。与几种典型边缘检测算子的比较实验表明,新算法可以较好地检测出图像边缘,相对于其他边缘检测算子,对于图像中的纹理等细节边缘有着更佳的检测效果;并且得益于BLMD图像多尺度分析的优势,较好地避免了因光照明暗等低频因素产生的假边缘,提取出的边缘更符合视觉上的主观检测。
基金Partially supported by National Natural Science Foundation and the Shanghai Qimingxing grant. # 97QA14040
文摘A modified version of the Cotte, Lions, Morel and Coil theory for image selective smoothing and edge detection is proposed. Comparing with their model, the most important advantage of this modification is that the convolution with Gaussian processes in the filtering process is replaced by solving an initial-boundary value problem for the heat equation, which simplifies the numerical scheme to some extent. Numerical experiments on natural images are presented for this model.
