【摘要】 在检测斜坡状边缘时,由于噪声的影响,经常出现单边缘多重响应现象.灰度形态学重建(grayscale recon struction)可以用于提取穹顶(dome).文章指出并证明了穹顶所具有的三个特性,并提出一种基于穹顶在梯度图中选取边缘的算法(DEG).根据指出的穹顶的特性,该算法具有非极大抑制(non maxima suppression, NMS)的效果;并且它仅选取出梯度图中相应于边缘的极大值点,在其附近由噪声引起的小的极大值点不会被选取出.算法的时间复杂度和内存要求同非极大抑制法相比没有明显的增加.实验结果显示,该算法抑制了单边缘多重响应现象,具有非极大抑制的效果,同时不会引起边缘的移位,适合用于对斜坡状边缘的检测.更多还原

【Abstract】 The phenomenon of multiple responses to a single edge often occurs when detecting noisy ramp edges from images. This paper first investigates grayscale reconstruction for dome extraction and concludes three properties of domes extracted by it. Based on the properties, a novel approach named DEG(Dome Extraction on Gradient image) is presented to select edges in gradient images. The approach extracts h domes in a gradient image along different directions row by row, and then selects the pixels whose value in h domes is equal to h as edge pixel candidates. At last, edge pixels are got by thresholding the candidates. Based on the concluded properties, it can be proved that above approach has the non maxima suppression(NMS) effect. In the gradient image, only the summits of main peaks corresponding to real edges are extracted, where as the little sharp maxima in edges’ vicinity caused by noises would not be chosen. Computation time and memory requirements of the approach are not more than those of the non maxima suppression approach obviously. Experiment results show that the approach suppresses the multiplicity of the response to a single ramp edge effectively without shifting edges that can be caused by smoothing methods, and gets the non maxima suppression effect simultaneity. The approach can be used in detecting ramp edges in noisy images.更多还原