Multiple Features Indexing in Image Retrieval Systems

In this paper we discuss the problem of indexing a large collection of images where each image is described by several features. Each feature de nes a multidimensional space where images are points, and the similarity between images is computed as the distance between points. We propose to approach multiple features indexing by maintaining a separate index structure for each feature, and merging the results during query processing. This approach, provides several advantages because allows di erent metrics to be used or di erent weights to be assigned to di erent features, leading to better retrieval quality. Some techniques for merging the sets retrieved by considering each feature separately are discussed, and experimental results on real as well as simulated data are shown.