基于局部PCA方向统计分析的Hough直线检测算法

A novel Hough transform line detection based on local PCA directions statistical analysis

为克服经典Hough变换本身存在的计算时间长、存储空间大的缺点,提出基于局部PCA方向统计分析的Hough直线检测算法。首先在边缘图像范围中选取合适大小的掩膜,通过局部PCA获得支持集内像素点的主元方向;侧重依据所有掩膜内主元方向信息的统计规律来约束Hough变换的极角选择范围,通过缩小参数选择范围大大缩短搜索时间。同时,每当一条直线检测完毕,当即把该直线上的点从原图像中除去,一方面可以避免直线间尤其是角度相近的直线间的影响,另外也减少下一轮极角子集范围内搜索的像素点数,提高运算效率。上述过程依此循环,直到检测出所有规定的直线。实验表明,所提出的算法计算精度高,运算时间短,并能提供直线段的完整描述。

To overcome the disadvantage of traditional Hough transform such as needing more storage and longer time, a method of Hough line detection based on local PCA （principle component analysis） directions statistical analysis is introduced. It first choose a mask with a suitable size in the edge image, the local PCA is performed to obtain the principal direction of the mask. The chosen polar angle in Hough transform is limited in a small range emphasizing particularly on the statistical disciplinarian of all principal directions, so it speed up the search time by reducing the range of θ. At the same time, the detected line is removed in image space when one line is extracted rightly, so it would avoid the impact among lines especially those whose angles are close and reduce the number ofpixels of next step to increase the efficiency. The process will continue till all the lines are detected. Experimental results demonstrate that the proposed algorithm has high detection accuracy, small computational requirements and the capability of providing complete line segment description.