Most commercially available image retrieval systems are so generic that they are not specialized to handle biological images and the feature domains associated with them. In molecular biology, neurobiology and cellular biology, for example, the recognition, classification and retrieval of distinct cellular features is a critically needed tool representing a computational problem that embodies the central challenges facing biological image database research. It often requires the consideration of expert/conceptual knowledge of images and the objects contained within such images.

This paper discusses the feasibility of developing a set of imaging algorithms that allows the user to train an imaging system/database to recognize biological objects of various sorts based on their own criteria. The image software builds a model of the selected objects by reiterative training, evolving the ability (i.e., the underlying rules) to recognize these objects. These objects are in turn archived into a growing database that builds upon the experience of multiple individuals that can be referred to as an object zoo. The user can search these zoos using newly acquired images based on similarly using more narrow or broadened criteria based on a new, semantic database framework called Semantic Objects.