Saliency-based Image Object Indexing and Retrieval

Abstract

We suggest a novel approach to combine visual saliency model and object recognition to provide a more semantic description of an image based on human attention priority. The idea is to index and retrieve semantically more relevant images utilizing human saliency. Based on that, we developed a content-based image indexing and retrieval system. The resultant indexing and retrieval system works, though there is room for improvement in performance. We suggest the reasons and the possibilities for further improvements to develop a practical CBIR system.