基于二值图像邻域加权的直线Hough变换

Hough transform for line detection based on neighborhood weighting in binary images

实际图像中普遍存在的纹理性区域,常常会由于所产生的大量特征点而形成Hough空间中的虚假峰值。为了抑制纹理性区域特征点造成的虚假峰值,提出了一种直接基于二值图像的邻域在线离线比加权方法,该方法通过考察投票特征点的邻域内在线与离线特征点的数量,来显著降低纹理性区域特征点的投票权重,从而使得Hough空间中真实直线对应峰值的显著性得到相对提升;提出了一种利用加权邻域内局部投票的该加权方法的高效实现,能够避免多次遍历邻域,减少运行时间。在实验图像库上的结果表明了新加权方法的有效性,反映了局部投票带来的运行速度的提高。

Textual regions commonly existing in real-world images often lead to false peaks in the Hough space gen- erated by large number of feature pixels in them. In order to suppress the false peaks produced by feature pixels in textual regions, a weighting method is proposed. The method is based on the online-offline ratio, which is directly computable from binary images. By investigating the number of online and offline feature pixels in the neighborhood of the voting pixel, the voting weights of textual region pixels are greatly reduced, and the relative significances of real line peaks in the Hough space is elevated. An efficient implementation of the weight computing scheme using local voting in the weighting neighborhood is given, which avoids multiple traversals over the neighborhood and thus speeds up the computation. The experimental results on an image base show the effectiveness of the proposed weighting method, and the promotion in execution speed realized by the local voting technique.