A flexible image retrieval using explicit visual instruction

We have developed a flexible image content retrieval system for silhouette images based on shape similarity. In our approach, a silhouette image is decomposed into primitives, each of which is translated into two signatures as indices of the image data. The two signatures are for attributes of each primitives and for relations between two of them, respectively. Queries are performed by the user´s drawing sketch and explicit instructions about attributes and relations with our developed sketch user interface. With this mechanism, the user can flexibly communicate detail requirements to the system and retrieve target images based on local/global shape features