基于奇异值分解的快速直线提取方法

Fast line detection algorithm based on singular value decomposition

矩阵的奇异值分解能够反映矩阵数据的分布特征。利用数字图像中直线的结构特征,定义了近垂直直线基元和近水平直线基元,根据基元结构对边缘检测和细化后得到的线段进行扫描,将近似共线的点分别归入到若干个直线支撑点集合。对每个直线支撑点集合进行奇异值分解,利用得到的特征向量计算出其对应的Hough参数空间投票单元,实现对线段层次上的特征的应用。实际图像处理结果表明,改进后的方法不仅能获得良好的直线检测结果,还能够大幅减少运算和存储方面的开销。

Line detection with Hough transform is a popular tool due to its robustness to noise and missing data.To overcome the huge computing burden and memory consumption in general Hough transform,a new line detection scheme which combined singular value decomposition was proposed.The singular value decomposition could reflect the distributed feature of matrix.The definition of vertical elementary of line and horizontal elementary of line were presented based on the characteristics of straight lines in digital image.After edge detection and thinning,a lot of clusters of approximately collinear pixels could be obtained by scanning the line segments using defined elementary of line.The position of vote unit in Hough space was computed using the singular value vector of each cluster.The experimental results show the improved method can not only accelerate the computing speed and save memory space,but also produce a much cleaner voting map and make the transform more robust.