On describing color and shape information in images

Abstract

With more and more audio-visual information being available over the Web, the problem of multi-media indexing, i.e. efficient ways of searching and locating multimedia data in response to queries, has become increasingly important. Efficient content-based access, however, needs a good description and representation of multimedia information. In the paper, we present descriptors for color and shape information in images that we proposed for MPEG-7. In particular, we describe two ways of capturing color content in images called Color Histogram and Region Color. These descriptors are suitable for a wide variety of applications requiring image-to-image matching and object-to-image matching. We also present descriptors for capturing shape information on objects and within images called affine structure and location hash tree. These shape descriptors enable a pose-invariant localization and recognition of 2d patterns. The utility of color and shape descriptors is demonstrated through experiments on the MPEG-7 dataset.