|
Shape classification is of fundamental importance in both computer vision and human vision. In this paper, the shape classification issue is solved through a pre-trained K-Nearest-Neighbor (KNN) classifier agaisnt a traslation-scale shape descriptor - regularized edge orientation curve (r-EOC). The r-EOC descriptor can also achieve rotation-invariant through implementing 1-D phase correlation. Thus with the pre-trained KNN classifier, the query shape is classified against the aggregate matching weight, which is composed of Eucledian distance of Translation-Rotation feature distance and the feature distance with smallest phase correlation. The demonstrated experimental result proves the effectiveness of the proposed method.
|
