Explorations in image databases
Ramesh Chandra Jain, Simone Santini · Medical Entomology and Zoology · 1998
This dissertation proposes Visual Computing and Perceptual search in databases as a new discipline, largely independent on the field of computer vision on which it rests. Computer vision tries to build anthropomorphic systems that carry out some specific perceptual task in a largely independent manner. Perceptual databases, on the other hand, are systems, whose operation only makes sense in connection with a user that guides the system and validates the results. This dissertation shows that this difference has important repercussions on all aspects of the design of an image database, from the data representation to the interface design. The second part of the dissertation proposes a mathematical model of image similarity suitable for application to social databases. The model is based on frames generated by discrete subgroup of a Lie group. The metric of Lie group is studied, as well as the metric of the color space in which images take values. This leads to the formulation of a theory in which image features are extremely general, and similarity criteria can be generated on the fly as a result of interaction with the user. The generality of the similarity model could overwhelm the user, and needs to be harnessed by a suitable interface design. The third part of this dissertation proposes guidelines for interface design and the design of an interface for a general purpose image database. The interface is based on the idea that the user should explore the image space looking for interesting regions, rather than asking a query and expecting a meaningful response.