Single chord-based corner detectors on planar curves

Detecting corner locations in the images plays a significant role in several computer vision applications such as motion detection, image registration, video tracking, image mosaicing, panorama stitching and object recognition. In this paper we have analyzed an existing state of art, Chord to Point Distance Accumulation (CPDA) corner detector and modified this detector in way that it uses a single chord instead of using three different chords. We named this detector as Single Chord CPDA (SCCPDA). We have also proposed a simple but effective new detector of detecting robust corner locations against different image transformations using cumulative distance calculation. The new detector has also used a single chord and we named it as Chord to Cumulative Sum Ratio (CCSR). A comprehensive performance evaluation has been performed by using Average Repeatability and Localization Error. We have found that the SCCPDA and CCSR detectors perform better than the original CPDA detector. Our experimental results show that the CCSR using simple cumulative calculation outperforms eight other existing contour based corner detectors in terms of repeatability and generates one of the lowest localization errors. In addition, the CCSR detector is also most efficient corner detector among other contour-based corner detectors.