Image Mining and Retrieval Using Hierarchical Support Vector Machines

Abstract

For some time now, image retrieval approaches have been developed that use low-level features, such as colour histograms, edge distributions and texture measures. What has been lacking in image retrieval approaches is the development of general methods for more structured object recognition. This paper describes in detail a general hierarchical image classifier approach, and illustrates the ease with which it can be trained to find objects in a scene. To further illustrate the wide capabilities of this approach, results from its application to particle picking in biology and Vietnamese art image retrieval are listed.