A Hierarchial Classification Technique for Semantics-Based Image Retrieval

Abstract

We present a new approach for improving image retrieval accuracy by integrating semantic concepts. First, images are represented according to different abstraction levels. At the lowest level, they are represented with visual features. At the upper level, they are represented with a set of very specific keywords. At the subsequent levels, they are represented with more general keywords. Second, visual content together with keywords are used to create a hierarchical index. A probabilistic classification approach is proposed, which allows to group similar images into the same class. Finally this index is exploited to define three retrieval mechanisms: the first is text-based, the second is content-based, and the third is a combination of both. Experiments show that our combination allows to nicely narrow the semantic gap encountered by most current image retrieval systems. Furthermore, we show that the proposed method helps reducing retrieval time and improving retrieval accuracy.