A New Adaptive Distance Computation Technique for Query-by-Multiple-Example System

Abstract

Query-by-one-example (QBOE) is the traditional way of querying in content-based image retrieval (CBIR) system. However, as some recent research points out, QBOE method cannot get accurate result because only one image is not sufficient to express its semantics of the intended query. Therefore, query-by-multiple-example (QBME) method is proposed and adopted, in which query images are divided into groups according to relevance to target image class. In order to maximize major features and minimize minor ones, previous researches have introduced adaptive distance computation in QBME. These methods optimize query result compared to QBOE, but still have some defects. This paper proposes a new adaptive distance computation technique for QBME, which achieves higher performance than previous methods.