Options
1998
Conference Paper
Titel
Interactive multisensor image exploitation. An approach to recognition starting from target objects
Abstract
Most common approaches to interactive object recognition in multisensor/multispectral imagery are sensor data driven. They address the problem of displaying images of multiple sensor sources in a manner adequate to the characteristics of the sensors. Fusion of sensed data is the topic of those concepts. This paper discusses a supplementing approach from the opposite end: the domain of target objects. Knowledge about the appearance of objects under various spectral conditions guides the image analyst through the interpretation process. Therefore, the basic concept of an "interactive recognition assistant" will be proposed. Starting from a set of candidate objects the image analyst is guided through a step-by-step interpretation process by getting indicated the respectively most significant features for efficient reduction of the candidate set. In the context of this approach we discuss the question of modeling and storing the multisensorial appearances of target objects as well as the p roblem of an adequate dynamic human-machine-interface that takes into account the mental model of human image interpretation.