Research directions in image database management

Abstract

The initial impetus for image databases originated with the image interpretation community. Most of the proposals from this community, however, were quite narrowly conceived and hence, after a brief flurry of activity in the late 1970's and early mid-1980's. interest in this activity decreased drastically. In our opinion. interest could not be sustained in this area due to its unsophisticated conception. At this time, the database community largely ignored such nonstandard applications due, we believe, to the then unsophisticated nature of the then current database management systems. It has only been since the development of various object-oriented approaches to database management that the field has expanded into these areas. 'In the last half of the 1980's. however, the situation had largely been reversed. The database community had expressed much interest in the development of nonstandard database management systems, including image databases, due, as mentioned above, to the development of the object-oriented paradigm as well as various data-driven approaches to iconic indexing. However, the interest of the image interpretation community had wavered. Only in the decade of the 1990's have the two communities been converging on a common conception of what an image database should be. This is due to the acceptance of the belief that image and textual information should be treated equally. Images should be able to be retrieved by content and should also be integral components of the query language. Thus, image interpretation should be an important component of any query processing strategy.There are many interesting problems in the field of image database management, including issues in data modeling, sensor data representation and interpretation. user interfaces, and query processing. The object-oriented paradigm has been and continues to be a great impetus to this work. We have seen major advances in image databases only since this paradigm has become accepted throughout the computer science community. However, in our opinion, this paradigm must further develop before image databases really comes into its own. In our opinion, methods should be treated as objects also and should be efficiently managed. We are only now starting to see image database systems where there are a fixed number of elementary methods and the user has the privilege (some would say the burden) of defining more complex methods in terms of more elementary ones. One of the reasons why we haven't seen much work in this area is quite apparenr it is difficult.Most researchers in image interpretation haven't been too concerned with the efficiency aspects of their work.Just developing a technique which works on real world data has been reward enough. However, there have been researchers, especially in the area of object recognition, who have been concerned with efficient methods for image interpretation. The field of image databases owes a lot to these people. While in the past, many of the issues addressed by such people were considered somewhat ...

<title>Multidimensional indexing for content-based image retrieval</title>

“…Research in information retrieval, database management, computer vision, and image processing is converging towards providing content-based access to image databases [9,10,16,18,19,20,21,22,23,24,25]. We report our recent research results on content-based image retrieval, based on an object-oriented image representation called blobworld developed by Carson, et al.…”

Effective invariant features for shape‐based image retrieval

“…Image shape descriptors, as used in existing CBIR systems, can be broadly categorized into two groups: contour-based and region-based descriptors. Contour-based shape descriptors exploit only information from the shape boundary, thereby ignoring potentially important information in the shape interior (Arbter, Snyder, Burkhart, & Hirzinger, 1990;Bober, 2001;Grosky et al, 1992;Jagadish, 1990;Sclaroff & Lifeng, 2001;Sclaroff & Pentland, 1995;Wu & Sheu, 1997). In regionbased methods, shape descriptors utilize information from the internal regions of the shape.…”