基于边缘切线方向连接的线段检测方法

Line Segments Detection Algorithm Based on Edge Tangent Direction Linking

关于图像线段检测优化问题,传统线段检测方法存在无法给出线段端点和长度、无法同时检测直线段和曲线段等不足,提出了一种从图像中检测直线段或者曲线段的方法,基本思路是假定图像中边缘点的边缘切线将邻域切分成两个区域,根据两个区域之间的均方差与均值的差异来寻找边缘切线方向,然后逐点连接切线倾角变化微小的相邻边缘像素得到线段。由于在相邻像素间移动模板时,只有模板边缘行列的像素对新的评估函数值产生影响,进一步给出了一个加速策略,使得计算连续边缘点切线方向的时间效率等同于枚举一个梯度算子。改进方法能够很好的判断出直线段或曲线的间断点,对于有较好对比度的灰度图像曲线段检测,传统的检测方法更快更准确。

As traditional line detection method such as Hough Transform cannot present the endpoints and length of line segments, and cannot detect straight line segments and curve segments at the same time, this paper presented an algorithm for detecting straight line segments and curve segments from images. The algorithm assumes that the tan- gent direction of edge pixels divides the local area into two parts. The differences of the mean square deviation and mean value between the two areas were used to find the tangent direction. The adjacent edge pixels that have small differences in their tangent directions were connected to form line segments. When the templates were moved among the pixels, only the marginal pixels affected the new assessment function value, and according to which an accelerating strategy was presented. With the accelerating strategy, the computing cost is merely equivalent to enumerate a spatial gradient operator. This algorithm can determine the disconnected points on lines well. Furthermore, for detecting curve segments in high contrast gray image, this algorithm has higher time effect and accuracy than traditional methods.