文摘
Browsing, searching and retrieving images from large databases based on low level color or texture visual features have been widely studied in recent years but are also often limited in terms of usefulness. In this paper, we propose a new framework that allows users to effectively browse and search in large image database based on their segmentation-based descriptive content and, more precisely, based on the geometrical layout and shapes of the different objects detected and segmented in the scene. This descriptive information, provided at a higher level of abstraction, can be a significant and complementary information which helps the user to browse through the collection in an intuitive and efficient manner. In addition, we study and discuss various ways and tools for efficiently clustering or for retrieving a specific subset or class of images in terms of segmentation-based descriptive content which can also be used to efficiently summarize the content of the image database. Experiments conducted on the Berkeley Segmentation Datasets show that this new framework can be effective in supporting image browsing and retrieval tasks.
