Shape Matching Using the Included Angle Histogram of Vectors

We propose a novel shape descriptor-Included Angle Histogram -for correspondence recovery of graphic vertex and shape-based object recognition. After detecting points local maximal curvature with and the center point of the contour, we construct vectors from the center point to the curvature points. Consequently the point descriptor can be obtained through computing the histograms of included-angles between this vector to the others. The descriptor at each point captures the distribution over corresponding points in the relative position, thus the global shape can be summarized by the combination of each local descriptor. It's invariant to geometrical transformation, eg. rotation, translation, scale and symmetry. The approach has been tested on the well known KIMIA99 database and MPEG-7 CE-Shape-1 database. The image retrieval results demonstrate that our descriptor is practical in shape matching. Moreover, compared to the other methods, it's more efficient in time-consuming.