Authors: Gerald Schaefer
Image collections are growing at a rapid rate, and efficient and effective techniques for managing these collections are highly sought after. Even though various tools and social networking platforms stimulate sharing images, locating images of interest in these vast repositories poses an interesting challenge. One of the main reasons for this is that most users do not annotate their image collections, and search is hence typically restricted to file attributes such as filename or date, which clearly has severe limitations. In this paper, we show how image databases can be indexed and queried based on content based image retrieval concepts. Content-based image retrieval extracts distinct features directly from the images and uses these to search for similar images. These image features typically describe colour, texture and/or shape properties and in this paper some of the basic descriptors that are used in this context are described. Following this, we then discuss more advanced topics including the retrieval of images in the compresseddomain (i.e., extracting image features without decompressing images) and the stability of colour features for image retrieval.