对偶点广义Hough变换算法的改进

Improvement of dual-point generalized Hough transform algorithm

对偶点广义Hough变换算法是通过寻找图像中像素梯度值相同的特征点对,并以该特征夹角作为R表索引,来完成表示目标边界信息的R表。但在检测不规则多边形等特殊图形时,由于符合条件的特征点对会有很多,因此在识别时会造成大量的无效干扰投票,甚至无法识别。从理论上分析了原算法产生虚假投票原因,提出了一种改进算法,以几何特征角作为新的索引。实验结果表明,改进是有效的,创建R表时分散了各索引项上的条目数,变换时减少了虚假投票和内存消耗,并提高了识别速度和识别率。

The dual-point generalized Hough transform found feature points with same gradient value in image and used characteristic angle for R-table index, then completed R-table representation ofobject boundary information. While it detected special irregular polygon graph, because the number of characteristic angle which accorded with condition is large, it generated many ineffective interference votes in object recognition, even identified failed. A theory dealing with the reasons of spurious votes in the prior algorithm is presented, and a modified algorithm is presented, and the geometrical characteristic angle is the new index. Experimental results show that the improvement is effective, the number of entries per index is decentralized in creating R-table, spurious votes and mcmory consumption is reduced in Hough transform, and the speed and accuracy of the recognition process is improved.