多尺度分析用于图像边缘的精确检测

Accurate Edge Detection Based on Multisolution Analysis

图像中边缘往往存在于一个较宽的尺度范围内,边缘的定位精确性是衡量一个边缘检测算法性能的重要依据。众多算法中,引起较大关注的是M-H的二阶导数过零点检测和Canny的一阶导数最大值检测,但都存在各自的不足。本文提出将线性权重函数(LWF)作为平滑函数,并利用小波理论的多尺度分析方法,对图像边缘进行检测,其在定位精确性、噪声抑制能力方面均有较好性能。

As intensity drastically changes at different scales within an image, the performance of an edge detector depends on its ability to accurately locate the real edge. Among those existing image segmentation techniques, Marr and Hildreth抯 LoG operator and Canny抯 optimal edge detector receives popular interests, but they also have shortcomings. The algorithm proposed in this paper uses LWF (line weight function) as smoothing filter, which can effectively imitate human visual receptive field. By means of multiscale analysis and the image edge is described with more accuracy and higher signal-to-noise-ratio.