Shape Corner Detection through Enhanced Curvature Properties

Image corner detection is a principal task for successful pattern recognition and identification. Finding corners of a shape usually involve two steps, firstly, shape segmentation and, secondly, locating corners in its boundary. This paper presents a robust method for detecting shape corner points using enhanced curvature properties. Numerous applications like image retrieval, object recognition, and motion tracking uses these corner points, which contain essential information about their location or position in a particular shape. Six existing shape detectors and descriptors, i.e., the Harris, SUSAN, Harris-Laplace, CSS, SIFT, and global and local curvature properties (GLCP), have been compared with the proposed shape detection method yielding its performance. The binary image dataset has been utilised as a benchmark to test the experiment. The experiment exhibits a significant number of true corners of the shape and outperforms the existing methods in terms of accuracy and reliability.