Querying the Content of Images in Material Science: An Integration of SQL and Map Algebra

Material scientists regularly acquire and analyze infrared images of deforming objects in their material tensile deformation, crack propagation, and fracture toughness tests. Although there are many image processing packages, none of them are available in a database integrated fashion. Material scientists typically select a set of image files satisfying given constraints by browsing a file directory/catalog, and then perform simple but labor-intensive contentquerying on the images. These simple queries take days to answer. A more serious problem is that material scientists are not equipped with the flexibility to query images across different time snapshots or materials to validate their research hypotheses. In this paper, we report about our work on helping material scientists accelerate their work. Specifically, we propose a database approach to solve the problem of storing images and querying the content of images. In particular, we (1) proposed to use map algebra operations to compose image content querying needed by material scientists; (2) developed an SQL integrated image data cartridge which implements a core set of map algebra operations needed by material scientists; (3) analyzed the query processing and evaluation challenges; and (4) empirically evaluated the performance of three approaches, namely a multi-dimensional array based approach, a relational table based approach, and a binary large object (BLOB) based approach on bulk loading and typical material science queries using both real and synthetic data.