A Novel Approach to Edge Selection on Gradient Images Based on Dome Extraction

Yi-Ke WANG,Wan-Chun DOU,Xi-Wen ZHANG,Shi-Jie CAI
DOI: https://doi.org/10.3321/j.issn:0254-4164.2005.02.016
2005-01-01
Jisuanji Xuebao/Chinese Journal of Computers
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.
What problem does this paper attempt to address?