Edge Orientation Spatiogram for Silhouette Image Retrieval

摘要

The edges provide important visual information. In order to enhance the description ability for image edges, a new descriptor is proposed by integrating the spatial distribution information of edge points with the edge orientation histogram (EOH), which is called edge orientation spatiogram (EOS). In order to invariant to scale transformation, translation, we utilize the polar radius to represent the spatial information for each edge point by using image centroid as origin. We test EOS on the MPEG7 database with several similarity measure methods and evaluate the performance in silhouette image retrieval scenario. The experimental results show that the edge feature descriptor EOS has a better performance than the traditional descriptor EOH.