Learning feature relevance and similarity metrics in image databases

Most of the current image retrieval systems use \one-shot " queries to a database to retrieve similar im-ages. Typically a K-NN (nearest neighbor) kind of al-gorithm is used where the weights of the features that are used to represent images remain xed (or manu-ally tweaked by the user) in the computation of a given similarity metric. However, neither all of the features are equally important for a given query nor a similarity metric is optimal for all kinds of images in a database. The manual adjustment of these weights and the se-lection of similarity metric are exhausting. Moreover, they require a very sophisticated user. In this paper we present a novel image retrieval system that contin-uously learns the weights of features and selects an ap-propriate similarity metric based on the user's feedback given as positive or negative image examples. Experi-mental results are presented that provide the objective evaluation of learning behavior of the system for image retrieval. 1