Reconstruction of Artifacts from Digital Image Repositories

Abstract

The U.S. Patent Office maintains an archive of cultural artifacts of both ornamental and functional designs. Design patents protect “any new, original, and ornamental design for an article of manufacture” [ 23 ] such as busts, statues, and the shape or configuration of any article of manufacture. The Patent Office archive originally contained both drawings and models as an effort to preserve a chronological history of the science and culture of the United States. This collection was devastated by two fires in 1836 and 1877, which destroyed or damaged more than 100,000 patent models. The Patent Office began divesting its collection in 1908, and Congress ultimately chose to dissolve the collection in 1926. A total of 10,000 of the original models are preserved in the collections of the Smithsonian National Museum of American History in Washington, D.C., and another 5,000 are preserved in the collections of the Hagley Museum and Library in Wilmington, Delaware. Although most of the physical artifacts no longer exist, documents describing the original physical models remain. These documents show each artifact as a collection of multiple black and white images depicting the artifact from different views. We propose an algorithm to reconstruct 3D point cloud models of the original artifacts from multiview images collected directly from the Patent Office archives. The use of 3D point cloud models allows both local and global geometric properties of the artifact to be described in a single vector representation. This vector representation of the artifact may be used individually or in combination with other descriptors for retrieval and classification tasks. For collections that are unclassified or where differing classification systems are used, these vector representations allow clustering of artifacts into meaningful groupings.