CBIR in Cultural Databases for Identification of Images: A Local-Descriptors Approach

Museums, archives and other cultural image collections, often are requested to identify an image whose metadata are inaccurate or missing. The task is further complicated because the query image may be an extract of the original, and present distortions, such as scale changes and noise. The majority of solutions proposed to this task are based on classic image signatures, such as the color histogram. Our approach, however, inherits from computer vision and is based on local descriptors. In this paper we describe our approach, explain the computer-vision method on which it is based and compared it to the Multiscale-CCV, an established scheme employed in a large scale practical system. We demonstrate experimentally the interest of our approach, which achieved a 99.2% success rate, against 61.0% for the Multiscale-CCV, in a database of photos, drawings and paintings.