A Fast Algorithm for Dominant Point Detection on Chain-Coded Contours
A Fast Algorithm for Dominant Point Detection on Chain-Coded Contours
复制标题
DOI：
10.1007/3-540-57233-3_33
复制
发表时间：
1993-09
期刊：
影响因子：
--
通讯作者：
Trond Melen;T. Ozanian
中科院分区：
文献类型：
--
作者：
Trond Melen;T. Ozanian
主导点检测是在高局部曲率的显著点处分割轮廓的任务。提出了一种新的，快速的自动检测这样的点的算法。对轮廓线分别进行了平滑处理和在任意方向上的微分处理。在曲率中具有局部极值的点被认为是候选优势点。对每个候选点计算由曲率及其导数组成的判别函数，并将其与阈值进行比较。平滑度和判别函数的阈值是算法的两个输入参数。计算成本在轮廓的长度上是线性的，并且对于合理尺寸的轮廓的实时视频（25- 30 Hz）处理来说足够低。该算法在许多真实轮廓上与人类观察者的性能相匹配，具有优异的定位精度，并且产生很少的伪点。
Dominant point detection is the task of segmenting a contour at distinguished points of high local curvature. A new, fast algorithm for automatic detection of such points is proposed. The contour is smoothed and differentiated in thexandydirection separately. Points with local extrema in the curvature are considered candidate dominant points. A discriminating function composed of the curvature and its derivative is computed for each candidate point and compared to a threshold. The degree of smoothing and the threshold on the discriminating function are the two input parameters to the algorithm. The computational cost is linear in the length of the contour and sufficiently low for real-time video (25–30Hz) processing of reasonable sized contours. The algorithm matches the performance of a human observer on many realistic contours, has excellent location accuracy and produces few spurious points.