Shape Recognition using Curve Segment Hausdorff Distance

A novel shape recognition approach (CsHD) combining the template and syntactic matching schemes is proposed in this paper to generalize the Hausdorff distance (HD) from matching two sets of points to matching two sets of lines/curves. The new approach retains the desirable features of HD. These include nonexplicit feature correspondence (natural allowance for portions of one shape to be compared with another) and simplicity in computation. In addition, a fast searching technique is employed to identify a list of likely matches. The proposed system has been applied to logo, palmprint and stationery recognitions with superior results. It has greatly out-performed the HD and is robust to noise, occlusion, skewing and minor broken line effect.