基于聚类的建筑物直线特征提取算法

建筑物图像中提取直线是视觉导航、特征识别等很多应用中的关键步骤。针对建筑物图像。提出了一种新的基于聚类的直线提取算法。该算法在Canny边缘的基础上，使用一个聚类算法将边缘分类，减少了因噪声形成的小短线对直线检测的误检或漏检现象；对于每一个分类使用直线识别准则，进一步剔除图像聚类中的伪直线；最后使用改进的Hough变换（standardHoughtransform）将直线提取出来。克服了传统Hough变换执行速度慢的缺陷。实验结果表明，文章提出的算法具有较快的运算速度，较高的直线提取准确性．较低的漏检和误检情况。其性能优于传统的Hough．变换和相位编组法。

An Extraction Algorithm for Line Feature of Architecture based on Cluster

Extracting lines in a digital image of architecture is a crucial step in many applications, such as visual navigation, feature recognition. In this paper, a new algorithm was proposed based on cluster for extracting straight lines from architecture image. Firstly, edges of cluster algorithm are detected based on Canny detector; Secondly, using an algorithm which follows image clustering categorizes edge to reduces the formation of small short lines due to noise on the line detection of false alarms or undetected; Thirdly, the rule of identifying line is used for eliminating false lines inside cluster image; Finally, straight lines was extracted by using a improved algorithm of Hough transformation to overcome the shortcomings of slow implementation on traditional Hough transformation. Experiment results showed that the proposed algorithm in the paper can detect lines accurately and quickly with low false detection and missing detection. In conclusion, the new algorithm performance is superior to the algorithms based on Hough transform and phase group on architecture images.