A New Method for Polygon Detection Based on Clustering

Polygon fitting is widely used in the field of image measurement. This paper proposes a polygon fitting method based on clustering algorithm. Image preprocessing is the first step, including bilateral filtering and binarization to extract a complete set of contour points. Secondly, the sliding window method is used to fit the point set in the window to a straight line to obtain the cyclic parameter set. We propose a variant of K-means algorithm. The algorithm only considers the transformation between adjacent clusters in each iteration, which ensures the order of the collection. In order to adapt to the cyclic point set, the boundary of the cluster will be sequentially changed after each iteration. Experimental results show that the algorithm can detect and fit polygons from contours with satisfactory accuracy and efficiency.