Progressive content-based retrieval using tree-structure graphs

An efficient hierarchical visual representation is proposed for progressive content-based image indexing and retrieval. More specifically, the image content is decomposed at different description levels, each of which represents the image at a different content scale. This is implemented by partitioning the image into objects structured in a hierarchical way. The lowest description level is obtained by applying a color segmentation algorithm to the image. On the other hand, the highest level is achieved by a semantic segmentation implemented in a semi-automatic framework. The spatial relationship of the object extracted, at a given description level, is described by a graph structure scheme. In particular, for each object (segment) a graph node is constructed and includes several features of that object. The graph links indicate the spatial relationship of the object with its neighbors. Following the hierarchical object decomposition, a tree-structure is created, providing efficient representation of the visual content.