Content-based retrieval from nonstationary image database

Image database systems should handle and retrieve images (based on their contents) from a large collection of images effectively and efficiently. A serious problem faced by these systems is the need to deal with the nonstationary database. In an image database system, image features are usually organized into an binary tree data structure, and to update the binary tree for a nonstationary database requires a large amount of computations. Here, we convert this difficult problem into a constrained optimization problem, and a scheme called IFDU (iteration-Free Data structure Updating technique) algorithm is proposed based on Lagrangian function to adapt the existing binary tree for a nonstationary database. Experimental results based on recall and precision reveal that our method provides a binary tree that is very close to the optimal one. According to the simulation result, our algorithm can maintain 91% correct-rate even when the number of new-coming images reaches 50% of the total number of images in the database.