用于线段特征提取的改进Hough变换

Improved Hough Transform Used in Line Segment Detection

线段是符号的一个具有旋转、平移和尺度不变的稳定性特征,正确提取符号的线段特征对于提高符号识别系统的识别率有很重要的意义。针对已有的基于Hough变换的线段提取算法的缺点,该文提出了一种用于线段特征提取的改进Hough变换算法。通过采用“多对一”映射;将Hough变换的投票过程和线段参数的检测过程融为一体;动态管理算法所需的临时存储空间等手段,使该算法具有较好的计算复杂度和空间复杂度。针对数字图像的量化特点,精心设计了用于检测在直线上点的条形区域,从而大大地降低了噪声对线段参数检测的影响,使该算法具有较好的检测性能和鲁棒性。实验表明,该文算法能正确提取出线段的端点坐标及其长度。

Line segment is a translation-,rotation-and scale-invariant feature of the symbol image.The extraction of the exact line segment feature is very significant to the recognition ratio of the system.Based on analyzing the drawbacks of the existing line segment detection methods based on Hough transform,an improved Hough transform used in line segment feature extraction is proposed in this paper,and can extract exactly the coordinates of the ending points and the length of the line segment.In our algorithm,the process of line segment detection is embedded in the process of Hough transform voting,the″many-to-one″mapping scheme and the dynamic storage strategy are adapted,so the improved algorithm needs less computational time and auxiliary space.To reduce the effect of the quantization of the digital image,we elaborately design a strip-shape detection region where all of the pixels belonging to the line are detected,but the noise points can hardly be detected.Theoretic analysis and experimental results show that this algorithm is better than existing algorithms in detection performance and robustness.