Content-based retrieval using local descriptors: Problems and issues from a database perspective

Most existing content-based image retrieval systems built above a very: lar ge database typically compute a single descriptor per image, based for exam ple on colour histograms. Therefore, these systems can only return images t hat are globally similar to the query image, hut cannot return images that contain some of the objects that are in the query. Recent image processing techniques, however, focused on line-grain image recognition to address the need of detecting similar objects in images. Fine-grain image recognition typically relies on computing many local descriptors per image. These techn iques obviously increase the recognition power of retrieval systems, but al so, raise neu problems in the design of fundamental lower-level functions s uch as indexes and secondary storage management. This paper addresses these problems: it shows that the three most efficient multi-dimensional indexin g techniques known today do not efficiently cope with the deep changes in t he retrieval process caused by the use of local descriptors. This paper als o identifies several research directions to investigate before being able t o build efficient image database systems supporting fine-grain recognition.