| Rapid developments in the Internet and multimedia applications allow us to access large amounts of image and video data. While significant progress has been made in digital data compression, content-based functionalities are still quite limited. For example, many existing techniques in content-based retrieval are based on global visual features extracted from the entire image. The objective of this work is to provide a better characterization of the media content by exploring the local visual information and building a region-based representation. Towards this goal, three major issues are addressed: region extraction, region description, and region search. For region extraction, a new algorithm called JSEG is developed for unsupervised segmentation of color-texture regions in the image. It provides robust results on a large set of real image data. Two spatio-temporal segmentation and tracking algorithms for video data are also developed, one suitable for rigid object motion and the other one for arbitrary non-rigid object motion. For region color, a compact Variable-bin Color Histogram descriptor is proposed. The descriptor is based on dominant color information, and results in image retrieval show that the proposed descriptor is compact and accurate compared to the traditional histogram method. For region search, an efficient database indexing scheme is designed for the color descriptor. It avoids the problems of high-dimensional indexing and provides fast retrieval. Prototype region-based retrieval systems are developed for searching image and video data using the proposed techniques. |
