Semantic browsing and retrieval in image libraries

In this paper, we address two main problems encountered in content-based image retrieval, namely the lack of image semantids that can be captured by extracting and indexing visual image features and the difficulty originating from the subjectivity and context dependency of user queries. This work proposes a new method for semantic browsing and retrieval of images by finding semantic coherence between words and image segments on three layers. The method is based on the matching of visual segment clusters with words on various levels of abstraction and is very promising for effective browsing and retrieval in large image databases. It supports various textual and/or visual query modes as well as both target- and category-type browsing and retrieval. Experiment conducted on a large set of natural images proved that step-by-step semantic inference on consecutive layers of image - word association helps to improve accuracy of retrieval and browsing.