Detecting Intrinsic Dissimilarities In Large Image Databases Through Skylines

In this paper we try to detect dissimilar images in image databases without defining a similarity ranking function by capturing the intrinsic dissimilarities of image descriptor vectors. To this end, we apply the skyline operation using their multi-dimensional descriptor vectors. We implemented a number of skyline methods, combined with four hashing state-of-the-art algorithms for data partitioning to create efficient indexing to secondary memory. We compared and evaluated their results by using two real image datasets to measure the performance and the effectiveness of the implemented algorithms. Detailed results show the efficiency and effectiveness of our approach.