Enhancing the Randomized Hough Transform with k-means clustering to detect mutually-occluded ellipses
Tinghui Zhou, Nikolaos P. Papanikolopoulos · 2011
In the attempts to resolve the problem of ellipse detection, the Randomized Hough Transform (RHT) serves as a powerful variant of the standard Hough transform that exploits the geometric properties of ellipses in order to speed up the detection process. Despite its simplicity and efficiency, the RHT performs poorly if the target ellipses are overlapped (or mutually-occluded) with each other. We present a novel method that utilizes k-means clustering to boost the performance of the RHT in detecting mutually-occluded ellipses, and test its effectiveness for both synthetic and real-world images. However, as a result of using k-means clustering, this method is susceptible to being stuck at a local optima.