A flexible content-based image retrieval system with combined scene description keyword

Proposes an image database with a fully automated keyword extraction function. Our approach can extract two different conceptual-level keywords from images. One is the "conceptual keyword", which is extracted by an image recognition technique using the "state transition model", which is a hierarchical model. The other keyword is the "scene description keyword", which is extracted by primitive parameters of color segments. We also propose the introduction of a "transition probability" to raise the retrieval accuracy (precision). Moreover, we evaluate the retrieval accuracy of this image database through a retrieval experiment using about 170 scenery images.